Académique Documents
Professionnel Documents
Culture Documents
Abstract
T HIS thesis covers the investigation into the use of Space Vector Modulation for the control of a 4-leg matrix converter, which is capable of
providing a 3-phase plus neutral supply from a standard balanced 3phase source. Traditional 3x3 matrix converters have limited use in this
application as they are only capable of supplying a balanced three-phase
load. It would be desirable to be able to power unbalanced and nonlinear loads, requiring that the converter provides a neutral connection.
As with voltage source inverters, this goal can be achieved by extending
the number of output legs in the matrix converter to four. In this thesis,
a new Space Vector Modulation technique is proposed for this 4-leg,
or 3x4, matrix converter. This technique is an extension of the method
currently in use on 3x3 matrix converters, and so it allows the de-coupled
control of both the input and output. The thesis then goes onto cover
the build of a demonstration converter, looking at the different aspects
which make up a converter, to nally go on to prove the theory, and a
set of results are presented to validate this.
ii
Acknowledgements
I would like to express my utmost thanks for my supervisors, Prof. P.
Wheeler and Prof. J. Clare, for both of their guidance and support, and
most importantly, for their utmost patience in me. I am sure there were
times when they, like me, gave up hope that this work would ever be
completed, but their willingness to help always meant that I could get
going going again.
Id do also like to thank all of the staff within the PEMC group at the
University of Nottingham who have helped me over the now many
years that this work has been in progress. Alongside this I would like
to specically thank Dr Lee Empringham and Dr Alan Watson for there
help at times with the build of the converter, who, along with the workshop technicians were always happy to help when I needed it and whose
help is much appreciated.
Id then like to thank my family and friends, they have been such a
immense support to me, more so than I think many of them know. They
have often picked me up and brushed me off when times were bad, and
so to them I owe a debt of gratitude. In particular though I would like to
say so many thanks to my wonderful partner Elizabeth Heslop, without
her help and support over the months and years that this has taken,
through the sad times and the happy, I am sure it would never have
been nished, and so for that I am eternally thankful. Id also like to say
thank you to my mother, whos always been a support to me, and has
really helped so many times.
And nally like to thank my sponsor, the Engineering and Physical Science Research Council(EPSRC), without whose support I would not have
been able to embark on this course of study, and I am sure they, as much
as anyone will be happy for me to complete.
iii
Contents
Acronyms
1
xvii
1
1.1
1.2
1.3
1.4
1.5
Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.6
2
Introduction
Thesis Overview . . . . . . . . . . . . . . . . . . . . . . . .
11
2.1
2.2
23
3.1
3.2
Alesina/Venturini Method . . . . . . . . . . . . . . . . . . . 26
3.3
3.4
4
Basic theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Commutation Methods
36
4.1
Overlap Commutation . . . . . . . . . . . . . . . . . . . . . 37
4.2
Dead-time Commutation . . . . . . . . . . . . . . . . . . . . 38
4.3
4-Step Commutation . . . . . . . . . . . . . . . . . . . . . . 39
4.4
C ONTENTS
4.5
5
5.2
5.3
5.4
5.4.2
5.5
5.6
5.7
5.8
5.9
90
6.2
4-Leg Inverter . . . . . . . . . . . . . . . . . . . . . . 91
4-Leg Matrix Converter . . . . . . . . . . . . . . . . 110
201
7.1
7.2
Devices . . . . . . . . . . . . . . . . . . . . . . . . . . 204
7.2.2
7.2.3
7.2.4
C ONTENTS
7.2.5
7.2.6
7.2.7
7.2.8
7.3
7.3.3
7.3.4
8
Matlab/PC . . . . . . . . . . . . . . . . . . . . . . . 221
Experimental Results
224
8.1
8.2
8.3
8.4
9
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
Conclusions
238
240
B IGBT Datasheet
241
247
References
253
vi
List of Figures
1.1
1.3
A 3-phase Cycloconverter . . . . . . . . . . . . . . . . . . .
1.4
1.5
A 4-Leg Inverter . . . . . . . . . . . . . . . . . . . . . . . . .
1.6
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2.9
1.2
3.2
3.3
3.4
L IST OF F IGURES
4.1
4.2
4.3
4.4
4.5
4.6
4.7
4.8
5.1
5.2
5.3
5.4
5.5
5.6
5.7
5.8
5.9
viii
L IST OF F IGURES
6.1
6.2
6.3
6.4
6.5
6.6
6.7
6.8
6.9
L IST OF F IGURES
6.22 Block Diagram of the MAST template for the Control Block 137
6.23 Circuit Diagram of the 4-Leg Matrix Converter for simulating the effects of the Input Filter . . . . . . . . . . . . . . 152
6.24 Plot of Control Block Internal Input Variables . . . . . . . . 156
6.25 Plot of Control Block Internal Output Variables . . . . . . . 157
6.26 Plot of Control Block Switching State Duty Cycles . . . . . 158
6.27 Plot of Demand and Output Phase-neutral Voltage for Leg a159
6.28 Plot of Output Phase-neutral Voltage for Leg a . . . . . . . 160
6.29 Plot of the FFT of the Output Phase-neutral Voltage on Leg a161
6.30 Plot of the 3 Output Phase-neutral Voltages at the load . . 162
6.31 Plot of the FFT of the Output Phase-neutral Voltage on Leg
a at the load . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
6.32 Plot of the Gate Control Signals . . . . . . . . . . . . . . . . 164
6.33 Plot showing the Switching of the Voltage on Leg a . . . . 165
6.34 Plot of Output Phase Voltages for Leg a . . . . . . . . . . . 166
6.35 Plot of the Input Current and Voltage for Supply Phase A . 167
6.36 Plot of the FFT of Input Current for Supply Phase A . . . . 168
6.37 Plot of the FFT of Input Current for Supply Phase A scaled
as a % of the fundamental . . . . . . . . . . . . . . . . . . . 169
6.38 Plot of the Output Currents showing an Unbalanced Load 170
6.39 Plot of the FFT of Output Phase-neutral Voltages with an
Unbalanced Load . . . . . . . . . . . . . . . . . . . . . . . . 171
6.40 Plot of the Input Current and Voltage for Supply Phase A
with an Unbalanced Load . . . . . . . . . . . . . . . . . . . 172
6.41 Plot of the FFT of Input Current for Supply Phase A with
an Unbalanced Supply . . . . . . . . . . . . . . . . . . . . . 173
6.42 Plot of FFT of the 3 Output Phase-neutral Voltages for
Unbalanced Demands . . . . . . . . . . . . . . . . . . . . . 174
6.43 Plot of the FFT of the Output Phase-neutral Voltage on Leg
a using Two Zero . . . . . . . . . . . . . . . . . . . . . . . . 175
6.44 Plot of the FFT of the Output Phase-neutral Voltage on Leg
a using Three Zero . . . . . . . . . . . . . . . . . . . . . . . 176
6.45 Plot of the Supply Voltage alongside the Sampled Voltage
showing the Processing Delay within the Converter . . . . 177
x
L IST OF F IGURES
6.46 Plot of the Supply Voltage alongside the Sampled Voltage
showing the Processing Delay Compensation . . . . . . . . 178
6.47 Plot of the FFT of the Output Voltage, Van, with and and
without compensation for the processing delay using the
Delta method . . . . . . . . . . . . . . . . . . . . . . . . . . 179
6.48 Plot of the noisy Input Voltage, VA, alongside the PLL
tracked fundamental frequency output used by the converter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
6.49 Plot of the Output Voltage, Van, and its Spectra showing
the effect of Noise when using the Delta method . . . . . . 181
6.50 Plot of the Output Voltage, Van, and its Spectra showing
the effect of Noise when using the PLL . . . . . . . . . . . . 182
6.51 Plot of the FFT of the Output Voltage, Van, with and without the truncation effect of the FPGA minimum step size . 183
6.52 Plot of the FFT of the Output Voltage, Van, showing the
original along with the output which is compensated for
the truncation effect of the FPGA minimum step size . . . 184
6.53 Plot of the Output Voltage, Van, and its Spectra, showing
the combined effect of the different types of compensation 185
6.54 Plot showing the switching sequence used for commutation going from a lower to a higher voltage input phase . . 186
6.55 Plot showing the switching sequence used for commutation going from a higher to a lower voltage input phase . . 187
6.56 Plot of the FFT of the Output Voltage, Van, comparing the
commutated output to the ideal . . . . . . . . . . . . . . . . 188
6.57 Plot of Input Phase A Supply Voltage and Current, along
with the voltage seen at the converter, when using a 1mH/33F
Input Filter with no PLL . . . . . . . . . . . . . . . . . . . . 189
6.58 Plot of Input Phase A Supply Voltage and Current, along
with the voltage seen at the converter, when using a 1mH/47F
Input Filter with no PLL . . . . . . . . . . . . . . . . . . . . 190
6.59 Plot of Input Phase A Supply Voltage and Current, along
with the voltage seen at the converter, when using a 1mH/33F
Input Filter using a PLL . . . . . . . . . . . . . . . . . . . . 191
6.60 Plot of Output Voltage, Van, along with its Spectra, when
using a 1mH/33F Input Filter using a PLL . . . . . . . . . 192
xi
L IST OF F IGURES
6.61 Plot of Input Phase A Supply Voltage and Current, along
with the voltage seen at the converter, with a changing
input voltage . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
6.62 Plot of Output Voltage, Van, along with its Spectra, with a
changing input voltage . . . . . . . . . . . . . . . . . . . . . 194
6.63 Plot of Input Phase A Supply Voltage and Current, along
with the voltage seen at the converter, with a changing
input frequency . . . . . . . . . . . . . . . . . . . . . . . . . 195
6.64 Plot of Output Voltage, Van, along with its Spectra, with a
changing input frequency . . . . . . . . . . . . . . . . . . . 196
6.65 Plot of Input Phase A Supply Voltage and Current, along
with the voltage seen at the converter, with a changing
input voltage and frequency . . . . . . . . . . . . . . . . . . 197
6.66 Plot of Output Voltage, Van, along with its Spectra, with a
changing input voltage and frequency . . . . . . . . . . . . 198
6.67 Plot of Input Phase A Supply Voltage and Current, along
with the voltage seen at the converter, with a changing
input voltage and frequency, driving an unbalanced load . 199
6.68 Plot of Output Voltage, Van, along with its Spectra, with a
changing input voltage and frequency, driving an unbalanced load . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
7.1
7.2
7.3
7.4
Circuit Diagram of the Gate Drive Circuit for a single Bidirectional Switch . . . . . . . . . . . . . . . . . . . . . . . . 206
7.5
7.6
7.7
7.8
Circuit Diagram of the Supply and Input Filter for the 4Leg Matrix Converter . . . . . . . . . . . . . . . . . . . . . . 212
7.9
Block Diagram for the DSP Software for the 4-Leg Matrix
Converter . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
7.10 Block Diagram showing the main components of the FPGA 218
xii
L IST OF F IGURES
7.11 The Main Control Interface for the 4-Leg Matrix Converter 222
7.12 The FGPA Registers Display Window . . . . . . . . . . . . 223
8.1
8.2
8.3
8.4
8.5
8.6
8.7
8.8
8.9
Plot of the 4-Leg Matrix Converter Input Current and Voltage for Phase A . . . . . . . . . . . . . . . . . . . . . . . . . 230
xiii
L IST OF F IGURES
8.18 Close-up of the FFT of the Output Phase-neutral Voltage
for Leg c across the Load for an Unbalanced Load . . . . . 236
8.19 Plot of the 4-Leg Matrix Converter Output Leg Currents
for an Unbalanced Load . . . . . . . . . . . . . . . . . . . . 236
xiv
List of Tables
2.1
2.2
2.3
2.4
3.1
3.2
3x3 Matrix Converter Stationary and Zero Switching Combinations transformed in the Plane . . . . . . . . . . . . 32
5.1
5.2
5.3
5.4
5.5
6.1
6.2
6.3
Switching state and Duty-Cycle selection for the 4-Leg Inverter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.4
6.5
6.6
6.7
xv
L IST OF TABLES
6.8
6.9
xvi
Acronyms
DSP digital signal processor.
FCC force-commutated cycloconverter.
FET eld effect transistor.
FFT fast Fourier transform.
FIFO First-In First-Out.
FPGA eld programmable gate array.
GPU ground power supply unit.
GTO gate turn-off thyristor.
GUI graphical user interface.
HPI Host Port Interface.
IGBT insulated gate bipolar transistor.
MEX Matlab executable.
NCC naturally commutated cycloconverter.
PCB printed circuit board.
PEMC Power Electronic and Machine Control.
PLL phase-locked loop.
PWM pulse-width modulation.
SCR silicon controlled rectier.
SVM space vector modulation.
xvii
A CRONYMS
UPS uninterruptible power supply.
USB Universal Serial Bus.
xviii
C HAPTER 1
Introduction
Over recent years there has been an increased interest in static power
converters, and this has led to a rapid improvement in their technology.
The use of such converters in various AC power supply applications has
also grown, and still continues to do so, with static power converters
now being a common part in many different systems, from uninterruptible power supplies (UPSs), power converters in submarines through to
ground power supply units (GPUs) for aircraft. One other typical application is for mobile power generation, and such a system is shown in Figure 1.1. This is an outline of a complete standalone system for providing
a three-phase power supply wherever it is needed using a variable speed
generator. The particular setup shown, uses the variable speed generator
connected to a power converter, which allows the engine driving the
generator to operate over a wide range of engines speeds and loads,
with the power converter then providing a xed frequency xed voltage
supply[1].
The power converter used in the scheme shown in Figure 1.1 could be a
standard 3-phase rectier/inverter setup, although this is a good setup
where a matrix converter could be used. There is a problem with both
of these types of converter, and it is that, while capable of generating
a balanced three-phase supply, they are not able to operate with unbalanced, non-linear, or single-phase loads. To enable the system to
generate an AC supply capable of supplying single-phase or unbalanced
three-phase loads the converter needs to be able to deal with the unbalanced neutral,or zero-sequence, current. One method for doing this
with an inverter is to increase the number of converter output legs to
include a neutral connection[2]. Extending this in the case of the matrix
converter would result in a 3x4, or four-leg, matrix converter as shown in
Figure 1.2. This shows a simplied version of Figure 1.1 with a four-leg
matrix converter attached to the generator.
1
C HAPTER 1: I NTRODUCTION
3-Phase
Generator
Bi-directional
Switch
Variable frequency
Variable output voltage
3-Phase plus neutral
C HAPTER 1: I NTRODUCTION
sets out to explore the design of a Space Vector Modulation technique
that is capable of controlling a 4-leg matrix converter, and then goes on
to detail the building and testing of such a converter.
The basis of the space vector modulation strategy for the 3x3 matrix
converter[3, 7] was derived from the techniques for both the controlled
rectier, and the inverter. This new technique will mirror this, and so
be derived from the modulation method used for a 3-phase rectier and
a four-leg inverter[12, 14]. This will use a space vector representation
in a three-dimensional space to describe the output voltage and
switching states[11, 12, 14]. This modulation technique will enable the
inverter, and thus the matrix converter, to provide any set of output lineto-neutral voltages, balanced or unbalanced, that may be required. This
three-dimensional approach will then be combined with the standard
two-dimensional SVM method for the input side of the matrix converter,
in a similar way to that performed for the 3x3 matrix converter[3, 4]. The
resulting SVM method will be a way of controlling the modulation of
the four-leg matrix converter so as to produce any set of line-to-neutral
voltages whilst also independently controlling the phase of the input
current waveforms.
1.1
C HAPTER 1: I NTRODUCTION
and are known as force-commutated cycloconverters (FCCs)
The naturally commutated cycloconverters use devices such as thyristors and SCRs to be able to control the point at which the device starts
conducting, thus, by choosing the correct phases from the input voltage,
an output voltage can be synthesised. However, due to the nature of the
devices they require the current in the device to reach zero before the
device can be turned off, so while it is possible to control the turn on of
each device, the turn off happens as the current naturally falls within the
device due to the changes in input voltage. Being able to control only
the device turn on means that the range output frequency is limited to
being below the input frequency[1517].
Va
VA
VB
VC
Vb
Vc
C HAPTER 1: I NTRODUCTION
to produce outputs at any frequency required, regardless of the input
frequency[5].
voltage was 23 [20] and then went on to show how this was possible with
their optimum method[6]. Lipo produced a similar result in 1989, while
also succeeding in building a working converter[21] to demonstrate this.
At around the same time, a number of papers by Huber and Borojevic
were published, which presented a new space vector modulation technique for the matrix converter [710]. This like the Alesina\Venturini
Optimum method, was able to produce the theoretical maximum output
voltage. Space Vector Modulation is a well known technique for controlling inverters, and it uses a technique which transforms the input and
output voltages and currents into a 2-dimensional plane where they can
then be easily manipulated. This is as opposed to the method Alesina
and Venturini used which is based around the use and theory of sine
waves.
Later on the space vector modulation technique was rened further by
Casadei[3, 4], and while more recent matrix converter congurations
5
C HAPTER 1: I NTRODUCTION
and modulation strategies have been proposed[22, 23], it is the technique
proposed by Casadei which this project seeks to extend onto the 4-Leg
Matrix Converter.
1.2
The four-leg inverter rst appeared in the early 1990s[24] when a solution was sought to allow the standard 3-leg inverter to drive unbalanced
and non-linear loads[2427]. Due to the nature of these loads they cause
an imbalance in the current drawn from each phase of the supply, which
then requires an extra neutral connection to deal with the zero sequence
current which then results. While the use of a split capacitors to provide
this neutral point was investigated[26, 28], this arrangement only works
well if the imbalance, or non-linearity, is small. Instead of the split capacitors, the 4-legged inverter uses an extra inverter leg to provide this
connection, and Figure 1.5 shows an example.
Vdc
Sn1
a pwm
n pwm
Vn
Sn2
n pwm
Sb1
Sa1
c pwm
b pwm
Vc
Vb
Va
Sa2
a pwm
Sc1
Sb2
b pwm
Sc2
c pwm
C HAPTER 1: I NTRODUCTION
conjunction with the SVM technique for matrix converters which was
proposed by Casadei[4].
1.3
The 4-leg Matrix Converter was, at the start of this project, a novel circuit
which had yet to see any work published about it, although this may not
be surprising with the majority of research effort in the matrix converter
eld being concentrated on the practical side to converter implementation, and the relative lack of appreciable applications.
1.4
One major application for this type of converter is as part of a eld power
supply, which is the type of system briey described above and shown in
7
C HAPTER 1: I NTRODUCTION
Figure 1.1, and which has since entered development by the University
of Nottingham[29, 30]. In this application it is able to demonstrate its
main advantage over other types of converter in that it is able to supply
power for not only balanced 3-phase sources, but also unbalanced and
single phase loads. So, along with the advantages that comes from being
a matrix converter, so it being compact and able to achieve high power
densities due to not requiring a DC link capacitor, means that it is highly
suited to this task.
Having the converter in this conguration, running directly from a generator also helps mitigate the biggest drawback of this converter, and
that is the quality of the input current when drawing an unbalanced
load. As will be demonstrated, once the load becomes unbalanced, due
to the direct nature of the converter with it having no energy storage
elements unlike a normal inverter, there is no way to buffer the supply
from the irregular power drawn, which, if the load is highly unbalanced
can lead to distortion of the supply voltage. However, with the standalone setup above these harmonic currents will be drawn from the
generator, and by careful speed control of the engine the worst effects
of this imbalance can be moderated.
One characteristic that this converter has, is the ability to individually
alter its output phase voltages, and frequencies, independently of one
another. While this might not immediately appear to have an application, it might be of some use when dealing with fault situations on the
load, where the converter would be able to balance the load power per
phase individually in order to not affect the supply, however such work
fell outside the scope of this project and so will need to be investigated
separately.
1.5
Objectives
The objective of this project was simple, to investigate the use of Space
Vector Modulation to enable the control a 4-Leg Matrix converter. To this
end it would have a number of outcomes:
Firstly, as this is a novel implementation of SVM on a new converter, it
would involve the understanding and analysis of the two similar types
of space vector modulation that previously existed, that of the 4-leg inverter and the matrix converter, and only from this point would it be
possible to extent both methods to cover the 4-leg matrix converter. Then
once the theory behind the operation of those converters was under8
C HAPTER 1: I NTRODUCTION
stood, it would then be possible to understand the rules which would
govern the behaviour of the 4-leg matrix converter. At this point it would
then be possible to derive the equations which would govern its operation.
The next objective would be to then produce a set of simulations which
would be able to accurately model the behaviour of the 4-leg matrix
converter using the derived equations. This would lead to a set of Matlab
programs which however this would rst require the implementation of
the two other converters to prove that the
The nal objective in this project is to build and run a working demonstration model of the 4-leg matrix converter. This would involve the
evolution of a pre-existing matrix converter design to add the extra output leg, and then require a new set of software to be able to operate and
control the converter.
1.6
Thesis Overview
C HAPTER 1: I NTRODUCTION
duty cycles.
Chapter 6 then builds on the work of the previous chapter, taking the derived equations and then initially simulates them mathematically within
Matlab, before moving on to perform a number of circuit based simulations within the Saber simulation software. These simulations will be
used to validate the derivation, while also becoming a good base for the
code which will be required to build an actual converter.
Chapter 7 describes the build of a demonstration converter based on
an existing 3x3 matrix converter design. This chapter briey looks at
each of the main parts that goes into building a converter, describing its
operation and going into detail where it is required.
Chapter 8 gives a summary of the results from testing the converter. It
shows how the output from the converter is very well matched with that
expected after the simulations, but then goes on to describe the issues
which the converter had during testing, which ultimately meant that
the converter failed and stopped testing short of the ideal, but still with
enough information to demonstrate that the concept worked well.
Chapter 9 then follows with the conclusions on the work presented here.
10
C HAPTER 2
Sb1
a pwm
Vn
b pwm
c pwm
a pwm
Vc
Vb
Va
Sa2
Sc1
Sb2
b pwm
Sc2
c pwm
a pwm
Vb
b pwm
Vc
c pwm
Va
Vc
0V
Vb
dc
- V2
Ts /2
Ts /2
Ts
a pwm
b pwm
c pwm
(b) 3-Leg 4-Wire Inverter PWM Timing
2.1
To overcome the limitations with the above circuit, instead of generating the neutral voltage using a split DC link the neutral voltage can be
generated by using a fourth inverter leg[24, 25, 27], leading to the actual
4-leg inverter as shown in Figure 2.3.
Vdc
Sn1
Sa1
a pwm
n pwm
Vn
Sn2
n pwm
Sb1
c pwm
b pwm
Vc
Vb
Va
Sa2
a pwm
Sc1
Sb2
b pwm
Sc2
c pwm
(2.1.1)
This constraint means that only 2 of the phase voltage variables are actually independent. Using this means that the three phase voltages can be
transformed into a 2-dimensional space, in this case the space using
(2.1.2)
Va
Vb
Vc
V
V
1 1
2
2
3
3
2
1
2
3
2
(2.1.2)
With a 3-leg inverter, each leg has only got two possible switching states
during operation, that is to switch that output leg to either the positive or
the negative of the DC link. With 3 legs, each with 2 possible switching
states, this gives 8 possible switching states that can be generated and
these are shown in Table 2.1 below.
Table 2.1: 3-Leg Inverter Switching Combinations
ppp
Va
Vb
Vc
nnn
pnn
ppn
npn
npp
nnp
pnp
1
2 Vdc
1
2 Vdc
1
2 Vdc
- 1 Vdc
2
1
- 2 Vdc
- 1 Vdc
2
1
2 Vdc
- 1 Vdc
2
- 1 Vdc
2
1
2 Vdc
1
2 Vdc
- 1 Vdc
2
- 1 Vdc
2
1
2 Vdc
- 1 Vdc
2
- 1 Vdc
2
1
2 Vdc
1
2 Vdc
- 1 Vdc
2
- 1 Vdc
2
1
2 Vdc
1
2 Vdc
- 1 Vdc
2
1
2 Vdc
ppp nnn
0
0
0
0
pnn
2
3 Vdc
0
ppn
1
3 Vdc
1
Vdc
3
npn
npp
nnp
pnp
1
2
1
1
- 3 Vdc - 3 Vdc - 3 Vdc
3 Vdc
1
1
1
Vdc
Vdc - Vdc
0
- 3
3
3
The vectors resulting from these 8 switching states can then be plotted
onto an Argand diagram, giving the well known result shown in Figure 2.4. This plot shows 6 space vectors placed equidistant from each
other around the origin, and these are created by the non-zero switching
states. The remaining two vectors, ppp and nnn, switch the same voltage
to each of the three output legs, and as such they are known as the zero
voltage states and do not produce a vector within the space.
14
pnp
ppn
npp
pnn
vd
npn
nnp
(2.1.3)
This means that all three phase voltages can now be considered to be
truly independent of each other, and so the 2-dimensional space does
not have enough degrees of freedom to be able to describe them fully. To
cater for the extra independent variable another degree of freedom needs
15
1
0
1 2 1
V
2
3
3
0
V = 0 2 2
1
1
1
3
V
2
2
2
2
Va
Vb
Vc
Vn
(2.1.4)
Now, due to inclusion of the extra neutral leg, instead of the 8 possible
switching combinations shown in Table 2.1 above for the 3-leg inverter,
there are now 16 different ones, which are shown in Table 2.3 below
Table 2.3: 4-Leg Inverter Switching Combinations
Van
Vbn
Vcn
Van
Vbn
Vcn
pppp
0
0
0
pppn
Vdc
Vdc
Vdc
nnnp
Vdc
Vdc
Vdc
nnnn
0
0
0
pnnp
0
Vdc
Vdc
pnnn
Vdc
0
0
ppnp
0
0
Vdc
ppnn
Vdc
Vdc
0
npnp
Vdc
0
Vdc
npnn
0
Vdc
0
nppp
Vdc
0
0
nppn
0
Vdc
Vdc
nnpp
Vdc
Vdc
0
nnpn
0
0
Vdc
pnpp
0
Vdc
0
pnpn
Vdc
0
Vdc
Transforming this into the space using (2.1.4) then gives the results
shown in Table 2.4.
Table 2.4: 4-Leg Inverter Switching Combinations in Space
pppp
0
0
nnnp
0
0
pnnp
2
3 Vdc
0
V
V
0
pppn
0
0
Vdc
Vdc
nnnn
0
0
0
2 Vdc
3
pnnn
2
3 Vdc
0
1
3 Vdc
V
V
V
ppnp
1
3 Vdc
1
Vdc
3
npnp
1
3 Vdc
1
Vdc
3
1 Vdc
3
ppnn
1
3 Vdc
1
Vdc
3
2
3 Vdc
2
3 Vdc
npnn
1
3 Vdc
1
Vdc
3
1
3 Vdc
nppp
2
3 Vdc
0
1
3 Vdc
nppn
2
3 Vdc
0
2
3 Vdc
nnpp
1
3 Vdc
1
3 Vdc
2
3 Vdc
nnpn
1
3 Vdc
1
3 Vdc
1
3 Vdc
pnpp
1
3 Vdc
1
3 Vdc
1 Vdc
3
pnpn
1
3 Vdc
1
3 Vdc
2
3 Vdc
The vectors resulting from the 14 non-zero switching states can then
be plotted in the space as shown in Figure 2.5. The remaining 2
vectors, pppp and nnnn, switch the same voltage to each of the four
output legs, and so they are known as the zero vectors as they do not
produce a voltage across the output. As with the vectors for the 3-leg
inverter shown in Figure 2.4, the vectors for the 4-leg inverter are all
positioned equidistantly from each other within the space, and the
shape they describe is a type of icositetrahedron (24-faced polyhedron).
16
Vdc
pppn
2Vdc
3
ppnn
nppn
Vdc
3
pnpn
npnn
pnnn
nnpn
ppnp
nppp
dc
V3
npnp
2Vdc
3
pnpp
pnnp
nnpn
nnnp
Vdc
Va = 240.00
(2.1.5)
Vb = 327.85
Vc = 87.84
Vn = 0
Converting this demand voltage into space using (2.1.4) produces a
vector, vd ,
V = 240.00
V = 240.00
V = 0
17
(2.1.6)
Vdc
3
pnnn
nnpn
0
vd
dc
V3
pnpp
d1
1
d2 =
Vdc
d3
1
2
1
2
3
2
3
2
23
3
2
V (240.00)
1 V (240.00)
V (0.00)
0
1
d z = 1 d1 d2 d3
(2.1.7)
(2.1.8)
Once the duty cycles are known, it is a simple case to produce the re18
2Vdc
3
pnpn
Vdc
3
pnnn
vd
dc
V3
pnpp
pppp
pnnn
pnpn
pnpp
a pwm
b pwm
c pwm
f pwm
tz
4
t1
2
t2
2
t3
2
tz
4
Ts
2
Ts
2
2.2
Alongside the Space Vector Modulation technique shown above, a complete carrier based implementation of the 4-leg inverter has been proposed by Kim and Sul[31] in 2004. This technique uses an offset voltage
19
Vn
Sa1
Vf
Sf2
Sb1
Sc1
Vc
Vb
Va
Sa2
Sb2
Sc2
Vdc Va f , Vb f , Vc f Vdc
(2.2.1)
Va f = Van Vf n
Vb f = Vbn Vf n
(2.2.2)
Vc f = Vcn Vf n
This allows the use of Vf n , the fourth inverter output leg, as an offset
voltage which can be manipulated to control the other 3 output leg voltages where
V
Vdc
Van , Vbn , Vcn dc
2
2
(2.2.3)
Vdc
V
Vf n dc
2
2
(2.2.4)
and
20
Vdc
V
Vmin Vf n dc Vmax
2
2
(2.2.5)
where
Vmin = min(Va f , Vb f , Vc f )
(2.2.6)
Vmid = mid(Va f , Vb f , Vc f )
(2.2.7)
Vmax = max(Va f , Vb f , Vc f )
(2.2.8)
(2.2.9)
as this would limit the possible offset voltages if the difference between
the maximum and minimum output voltages is equal to or greater than
the DC link voltage.
The optimum level for Vf n is then be calculated as
Vf n = mid
V
Vmax + Vmin
Vmax
, min ,
2
2
2
(2.2.10)
Ts
2
Ts
Tb =
2
Ts
Tc =
2
Ts
Tf =
2
Ta =
Van
Ts
Vdc
V
+ bn Ts
Vdc
Vcn
+
Ts
Vdc
Vf n
+
Ts
Vdc
(2.2.11)
(2.2.12)
As demonstrated by Kim and Sul[31], this method gives equivalent performance to the 3-dimensional Space Vector Modulation method as described above.
21
Van
Vaf
a pwm
Vbn
Vbf
b pwm
Vcn
Vcf
c pwm
Calc Vfn
Vfn
f pwm
Van
Vcn
Vfn
Vbn
dc
- V2
Ts /2
Ts /2
Ts
a pwm
b pwm
c pwm
f pwm
(b) 4-Leg 4-Wire Inverter Carrier-based PWM Timing
22
C HAPTER 3
23
3.1
Basic theory
If we dene the input and output voltages of the converter as the vectors (3.1.1) and (3.1.2) respectively
VA (t)
Vi (t) = VB (t)
VC (t)
V (t)
a
Vo (t) = Vb (t)
Vc (t)
(3.1.1)
(3.1.2)
(3.1.3)
(3.1.4)
n= A,B,C
sna (t) =
snb (t) =
n= A,B,C
snc (t) = 1
(3.1.5)
n= A,B,C
(3.1.6)
tnp
Tsamp
(3.1.7)
Taking this set of modulation indexes the modulation matrix M(t) can
be constructed such that
Vo (t) = M(t) Vi (t)
(3.1.8)
(3.1.9)
and
m Aa (t) m Ba (t) mCa (t)
VA (t)
Va (t)
Vb (t) = m Ab (t) m Bb (t) mCb (t) VB (t)
m Ac (t) m Bc (t) mCc (t)
VC (t)
Vc (t)
(3.1.10)
m Aa (t) m Ab (t) m Ac (t)
Ia ( t )
I A (t)
IB (t) = m Ba (t) m Bb (t) m Bc (t) Ib (t)
mCa (t) mCb (t) mCc (t)
Ic (t)
IC (t)
(3.1.11)
and
Once again, to satisfy the requirement that no input phase is short circuited and no output leg is open circuit
mna (t) =
n= A,B,C
mnb (t) =
n= A,B,C
mnc (t) = 1
(3.1.12)
n= A,B,C
where 0 mnp 1.
These are the basic starting assumptions on which the different modulation strategies as based.
25
3.2
Alesina/Venturini Method
vi cos(i t)
Vi (t) = vi cos(i t + 2 )
3
vi cos(i t + 4 )
3
io cos(o t + o )
Io (t) = io cos(o t + 2 + o )
3
io cos(o t + 4 + o )
3
(3.2.1)
(3.2.2)
vo cos(o t)
Vo (t) = vo cos(o t + 2 )
3
vo cos(o t + 4 )
3
ii cos(i t + i )
2
Ii (t) = ii cos(i t + 3 + i )
4
ii cos(i t + 3 + i )
(3.2.3)
(3.2.4)
The solution to this the generalised equation for the modulation function
M(t) as
1 + 2qCS(0)
1
4
M ( t ) = 3 1
1 + 2qCS( 3 )
2
1 + 2qCS( 3 )
1 + 2qCA(0)
1
+ 3 2 1 + 2qCS( 2 )
3
1 + 2qCS( 4 )
3
2
1 + 2qCS( 3 )
1 + 2qCS(0)
1 + 2qCS( 4 )
3
2
1 + 2qCS( 3 )
1 + 2qCS( 4 )
3
1 + 2qCS(0)
26
1 + 2qCS( 4 )
3
2
1 + 2qCS( 3 )
1 + 2qCS(0)
4
1 + 2qCS( 3 )
1 + 2qCS(0)
1 + 2qCS( 2 )
3
(3.2.5)
with
1 0
2 0
1
0 q 2,
Solving the above equation gives the required modulation indices for
each of the switches, and so can be used to directly drive the converter.
There is however one problem with this method, which is shown in the
last equation, and that is the maximum voltage transfer ratio q is only
0.5, meaning that the output voltage can only be a maximum of half the
input.
3.3
To overcome the low voltage transfer ratio found in the original Alesina/Venturini method it was proposed that the transfer ratio could be increased
by adding a proportion of the third harmonic into the output frequency,
and due to the balanced nature of the loads being driven, this third
harmonic would boost the output voltage yet not feed back into the
supply side current. This method was formalised in 1989 by Alesina
and Venturini which became known as the Optimum Method[6] and
27
Solving for the modulation matrix M(t) then gives the following from[6]:
Z ()(t) = cos (o + i )t +
(3.3.2)
and
1
3
1
m ( x1 , x2 , x3 , x4 , x5 , x6 ) = 1 +
p Z1 ( x1 ) + Z1 1 ( x2 )
3
2
1
1 1
Z3 ( x3 ) Z3 1( x4 )
6
6
1
7
4
2
+ sgn( p) Z0 ( x5 ) + Z0 ( x6 )
6 3
6 3
1
+ a1 Z1 ( x1 ) + a2 Z1 1( x2 )
(3.3.3)
where
tan(vi )
tan(v0 )
a = 2| | v o
vi
2 vo a
vi
p=
3
a1 = a and a2 = 0 if < 0
a1 = 0 and a2 = a if > 0
a1 = a2 = 0 if = 0,
then
3.4
While the Space Vector representation had previously been used for inverter control it was not until Huber and Borojevic in 1989[710] where
this method was proposed for use with matrix converters. This was then
extended with a paper by Casadei at al in 2002[3, 4], giving a full Space
Vector Modulation strategy that controlled both the output voltage and
input power factor.
This space vector modulation approach to matrix converter modulation
is all based on the space vector representation of the input and output
voltages and currents at any one instant. For a 3-phase set of line-toneutral voltages this is given by
V(t) =
2
v a + avb + a2 vc
3
(3.4.1)
where
a = ej
2
3
(3.4.2)
Plotting the three vectors that make up V(t) on an Argand diagram gives
a set of 3 vectors spaced evenly 120 apart due to the a and a2 terms.
Plotting V(t) on the same diagram then produces a vector of constant
length, vo rotating about the origin at the output frequency o and with
a phase angle of o , as shown in Figure 3.1.
The basis of the Space Vector Modulation technique is that the output
voltage Vo (t), which is expressed in the above space vector form (3.4.1),
can be generated by switching between adjacent space vectors and producing a time averaged value over a switching period that is equal to
the required output space vector. This is the same technique as used in
conventional inverters but by adding in the control of the input current it
allows complete control over the operation of a matrix converter. However, instead of the 8 possible switching states of the 3-phase inverter,
29
vb
Vo
o
o
va
vc
Figure 3.1: Base set of SVM Vectors for a Balanced 3-Phase Set plotted in
the plane
there are now 27 possible switching states. This is because instead of just
switching the DC link onto the outputs, the converter has the three input
phases to use. Now, unlike with the inverter, not all of the switching
states are usable and they fall in the following three groups
Group I : Where all 3 input phases are switched to the output,
thus producing a constant amplitude vector which rotates about
the origin at the input frequency.
Group II : Where any 2 input phases are switched to the output,
meaning that two output legs are connected to the same phase.
This gives a vector of a xed angular displacement but with varying amplitude.
Group III : Where only a single phase is switched to all 3 outputs,
meaning that there is no difference in voltage between the output
legs. This gives a vector of zero length placed at the origin.
Of these three groups, only Groups II and III, which are shown in Table 3.1 are required in this method, with the rotating Group I switching
states not being used as they cannot readily be employed to generate the
required output voltages.
30
Leg
a
A
B
B
C
C
A
B
A
C
B
A
C
B
A
C
B
A
C
A
B
C
Leg
b
B
A
C
B
A
C
A
B
B
C
C
A
B
A
C
B
A
C
A
B
C
Leg
c
B
A
C
B
A
C
B
A
C
B
A
C
A
B
B
C
C
A
A
B
C
Vab
Vbc
Vca
VAB
VAB
VBC
VBC
VCA
VCA
VAB
VAB
VBC
VBC
VCA
VCA
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
VAB
VAB
VBC
VBC
VCA
VCA
VAB
VAB
VBC
VBC
VCA
VCA
0
0
0
VAB
VAB
VBC
VBC
VCA
VCA
0
0
0
0
0
0
VAB
VAB
VBC
VBC
VCA
VCA
0
0
0
In the same way as was done for the switching states for the 4-leg inverter, the switching states in Table 3.1 are then transformed into the
plane using the transform
V
V
2
3
1 1
2
3
2
1
2
3
2
Va
Vb
Vc
(3.4.3)
31
vo
ii
2
3 VAB
2 VAB
3
2
3 VBC
2 VBC
3
2
3 VCA
2 VCA
3
2
3 VAB
2 VAB
3
2
3 VBC
2 VBC
3
2
3
2
3
2
3
2
3
2
3
2
3
2
3 VCA
2 VCA
3
4
3
4
3
4
3
4
3
4
3
4
3
2
ia
3
ia
23
2
ia
3
ia
23
2
ia
3
ia
23
2
ib
3
ib
23
2
ib
3
ib
23
2
ib
3
2
3 i b
2
ic
3
2
3 i c
2
ic
3
2
3 i c
2
ic
3
2
3 i c
01
02
03
No.
+1
1
+2
2
+3
3
+4
4
+5
5
+6
6
+7
7
+8
8
+9
2
3 VCA
2 VCA
3
2
3 VAB
2 VAB
3
2
3 VBC
2 VBC
3
0
0
0
0
0
2
7
6
7
6
2
7
6
7
6
2
7
6
7
6
6 switching states that can generate this output voltage vector depending on the sign of the input voltage. However, each of these switching
states generates a different input current vector, as shown in Figures 3.2
and 3.3, and it is this link between the input current and the output
voltage that allows the selection of the switching states, and thus control
over the converter. Due to the number of possible switching states it
is always possible to nd a set of 4 states that can generate the output
voltage required while also controlling the phase of the input current.
This method is set out by Casadei [3, 4] and is generated as follows.
32
4
5,
6
vo
1, 2, 3
8,
I
I
vo = vo I + vo I I I
j [ ( K v 1) + ]
2
3
3
e
= vo cos o
3
3
I
IV
vo = vo I I I I I + vo IV
2
j [ ( K v 1) ]
3
= vo cos o +
e
3
3
(3.4.4)
(3.4.5)
33
2, 5, 8
vi
ii
i
i
,
6
4,
ii
,+
+1
vo
+7
,+
+1, +3
Figure 3.4: Input Current and Output Voltage Sectors, showing the
required switching states
(3.4.6)
(3.4.7)
Solving equations (3.4.4) to (3.4.7) with respect to the duty cycles then
gives us
34
2 cos o cos i
3
q
= (1)
cos i
3
2 cos o 3 cos i +
3
I I = (1)Kv +Ki q
cos i
3
2 cos o + cos i
3
3
I I I = (1)Kv +Ki q
cos i
3
cos o + cos i +
IV
K v + Ki +1 2
3
q
= (1)
cos i
3
I
K v + Ki +1
(3.4.8)
(3.4.9)
(3.4.10)
(3.4.11)
| I | + | I I | + | I I I | + | IV | 1
(3.4.12)
and from here it is possible to the calculate the maximum possible transfer ratio, q, as
3 | cos i |
2 cos i cos o
(3.4.13)
These equations, with those described in Section 2.1, are going to form
the basis of the derivation of space vector modulation of the 4-leg matrix
converter.
35
C HAPTER 4
Commutation Methods
As has been stated several times, there are two fundamental rules for
the operation of a matrix converter that arise from the voltage-stiff input
and current-stiff output characteristics. The rst is that no two input
phases can be shorted together and the second is that no output phase
can go open circuit. Because of these restrictions, and due to the bidirectional nature of the switches that are needed for the converter to
operate, it means that there is a problem with the two most commonly
used commutation strategies for the inverter, Overlap and Dead-time.
This chapter will rst look at the Overlap and Dead-time strategies, and
explain why neither of these is really suitable for the needs of the matrix
converter, and will then go on to look at a number of other types which
do not have the drawbacks of the rst two.
S Aa
S Aa2
S Aa1
SBa
SBa2
Load
SBa1
4.1
Overlap Commutation
SAa
SBa
S Aa1
S Aa2
SBa1
SBa2
td
(a) Overlap Current Commutation Timing Diagram
a
1
1
0
0
b
0
0
1
1
1
1
1
1
S Aa1
S Aa2
SBa1
SBa2
4.2
Dead-time Commutation
SAa
SBa
S Aa1
S Aa2
SBa1
SBa2
td
(a) Deadtime Current Commutation Timing Diagram
a
1
1
0
0
b
0
0
1
1
0
0
0
0
S Aa1
S Aa2
SBa1
SBa2
4.3
4-Step Commutation
The 4-Step commutation process was one of the rst methods proposed
that allowed the matrix converter to be successfully switched between
states without violating either of the two rules dened above for their
operation. However, for this approach to work it requires a bi-directional
switch which can independently control the conduction path in each
direction, this is the example that is shown in Figure 4.1 above. As its
name suggests, unlike the previous two methods this method uses four
individual steps and this sequence is demonstrated in Figure 4.4
The rst step is for the non-conducting device in the outgoing switch to
be turned off( a). Then in the incoming switch, the device which will be
conducting is switched on(b). This can happen without creating a short
circuit between the two input phases because both switches are only
being turned on in one direction only, and both in the same direction
as the current ow, so the short circuit is not possible due to the series
diodes in the bi-directional switch. At this point it is then possible to
turn off the conducting device in the outgoing switch(c), turning off that
bi-directional switch entirely. The nal step, once the outgoing device
is completely turned off, is then for the non-conducting device in the
incoming switch to be turned on(d).
Unlike the 2-step commutation processes above, which only need to
know which switches are involved, this technique also needs to know
the current direction within the switch. This is important as it is the current direction within the bi-directional switch that denes which order
the devices are switched, and this is simple to achieve at higher currents
using hall effect sensors on the output legs. However there is a problem
using the hall effect sensors at very low currents, but at these current
levels, while not ideal, it would be possible to switch to a dead-time
39
SAa
SBa
S Aa1
S Aa2
SBa1
SBa2
td
(a) Four-Step Current Commutation Timing Diagram
b
1
0
0
0
IL < 0
c
1
0
1
0
d
0
0
1
0
1
1
0
0
IL > 0
0
0
1
1
0
1
0
0
0
1
0
1
S Aa1
S Aa2
SBa1
SBa2
0
0
0
1
4.4
As can be seen in the four-step commutation process above, at the required switching instant the rst and last steps do not really appear to
achieve much as neither of the devices are conducting, and they just slow
down the overall switching process as the incoming device only gets
turned on td after the switching process has started. To allow for this the
threshold 2-step commutation process was derived. It is essentially as
the four-step process except that it only ever turns on the conducting ele40
SAa
SBa
S Aa1
S Aa2
SBa1
SBa2
td
(a) Threshold 2-step Current Commutation Timing Diagram
a
1
0
0
0
IL < 0
IL 0
IL > 0
b
1
0
1
0
0
0
1
0
1
1
0
0
0
0
1
1
0
1
0
0
0
1
0
1
S Aa1
S Aa2
SBa1
SBa2
0
0
0
1
IL < 0
IL 0
IL > 0
1
1
0
0
b
1
0
1
0
0
0
1
0
0
0
1
1
0
0
0
0
0
1
0
0
0
1
0
1
S Aa1
S Aa2
SBa1
SBa2
0
0
0
1
4.5
vy
vx
43
SAa
SBa
S Aa1
S Aa2
SBa1
SBa2
td
(a) Voltage Sensed Current Commutation Timing Diagram
a
1
0
0
0
IL < 0
IL 0
IL > 0
b
1
0
1
0
0
0
1
0
0
0
0
0
0
0
0
0
0
1
0
0
0
1
0
1
S Aa1
S Aa2
SBa1
SBa2
0
0
0
1
44
C HAPTER 5
45
5.1
Looking at the diagram in Figure 1.6, it shows the basic layout of the
4-leg matrix converter. From this it can be seen that while there are 4
output legs to the converter, there are only 3 input legs, so while the
output is capable of generating an unbalanced set of phase voltages, the
input must form a balanced three phase set as dened by
VA + VB + VC = 0
(5.1.1)
VA = Vi cos (i )
(5.1.2)
2
3
4
VC = Vi cos i
3
VB = Vi cos i
(5.1.3)
(5.1.4)
where Vi is the phase input voltage, and i is the phase of the input
voltage at a particular instant in time. An example set of input voltages
are also shown in Figure 5.1(a).
In the same way as for the 3x3 matrix converter, this set of input voltages
can be represented as a space vector by using the following well known
-transform
V
V
2
3
1
0
1
2
3
2
1
2
23
VA
VB
VC
(5.1.5)
This equation (5.1.5) takes the three input phase voltages and calculates
the values V and V , which can be plotted as a vector vi in the plane.
Figure 5.1(c) shows an example where vi has been plotted for the input
voltages shown in Figure 5.1(b).
As the input is a balanced 3-phase set this vector will be of a constant
amplitude, |vi |, and rotate around the origin at the input frequency, i .
Using equations (5.1.1) and (5.1.5) it is possible to write down three
primary voltage space vectors which relate to the three input phase voltages, and there are labelled as Av , Bv , Cv , such that
46
400
300
VA
200
100
VC
Vin 0
-100
-200
-300
-400
VB
0
10 12 14 16 18 20
time (ms)
V (240)
V (240)
2
3
1
2
3
2
1
0
1
2
23
VA (240)
VB (328)
VC (88)
(b) Transform
Bv
Av
VA
VC
VB
Vi
Cv
Figure 5.1: Input space showing the 3 primary input voltage vectors
47
Av =
2
3
1
0
Bv =
2
3
1
2
3
2
Cv =
2
3
1
2
23
(5.1.6)
These three vectors can be seen plotted in Figure 5.1(c), and are the three
vectors spaced evenly 120 about the origin.
5.2
In exactly the same way as for the input voltage plane shown above, the
input currents, I A , IB and IC , can also be transformed into the space
using the same transform as the input voltage, so
I
I
2
3
1
2
3
2
1
2
23
IA
IB
IC
(5.2.1)
As for the voltages above, the values I and I can be plotted to produce
an input current space vector, ii , in the space whose magnitude, |ii |,
and phase, i , are determined by the operation of the converter and by
the load. However, the converter is being designed to achieve a unity
input displacement factor which means that the fundamental frequency
of the input current should be in phase with the fundamental frequency
of the input voltage. In this case the phase angle of the input current
space vector will always match that of the input voltage vector, and so
will rotate around the origin at the input voltage frequency, i .
Therefore, as the input voltage vector was formed by a balanced 3-phase
set, and the input current vector rotates in phase with the resultant voltage vector, then then at any one instant the input currents must also form
a balanced 3-phase set such that
I A + IB + IC = 0
(5.2.2)
From equation (5.2.2) and using equation (5.2.1), the three primary current space vectors, Ai , Bi , Ci relating to the current in each input phase
can be written down as
Ai =
2
3
1
0
Bi =
2
3
1
2
3
2
48
Ci =
2
3
1
2
23
(5.2.3)
Bi
IA
i
IC
IB
Ai
ii
Ci
Figure 5.2: Input space showing the 3 primary input current vectors
Figure 5.2 shows the input current plane, with these three primary
current vectors plotted, these are the three vectors spaced evenly 120
about the origin, and are the same as for the input voltage. Also shown
is the direction of the desired input current vector ii for the example
shown in Figure 5.1.
5.3
The 3x3 matrix converter produces 3 different output voltages, but because these voltages are all referenced to each another, and form a balanced set, they are constrained by the following equation
Va + Vb + Vc = 0
(5.3.1)
So while there are three different variables, Va , Vb and Vc , they are all related to one another and so this actually only gives 2 degrees of freedom.
This means that it is only possible to independently set 2 out of 3 of the
output voltages, with the third being set by the solution to equation 5.3.1.
With the 4-leg matrix converter, which is able to produce 4 different output voltages, once again all the voltages are referenced to one another,
but are now constrained by
49
Va + Vb + Vc + Vn = 0
(5.3.2)
(5.3.3)
X = 0
3
X
2 2
1
2
3
2
1
2 2
1
2
3
2
2 2
Va
V
b
0
Vc
2 2
Vn
0
(5.3.4)
(5.3.5)
Using the transform in equation (5.3.4) the four space vectors repre-
50
400
300
200
100
0
-100
-200
-300
-400
10 12 14 16 18 20
X (120)
X (120) =
X (0)
2
3
2 2
1
2
3
2
1
2 2
V (120)
a
0
Vb (164)
0
Vc (44)
3
2 2
Vn (0)
1
2
23
1
2 2
(b) Transform
c
Vc
Va
Vb
Vd
n
(c) Output Space showing the example from (b)
51
2
a= 3 0 b=
2 2
2
3
1
2
3
2
1
2 2
c=
2
3
1
2
23
1
2 2
n=
2
3
0
0
3
22
(5.3.6)
These are the four primary output vectors, and are also plotted on Figure 5.3(c) where it can be seen that, in a similar way to the input voltage
space vectors before, these are all equidistant from one another within
the space.
It should be noted that between the two papers which were used as
the source for the 3-dimensional space vector modulation that is used
in this derivation, there are a couple of differences with this transform
matrix. The rst being that Ryan, Lorenz and De Doncker[12] use the
Tdqo notation more commonly used for a rotating frame of reference,
while Zhang[14] uses the static notation. The notation which
was chosen for this derivation as the vector space does not move, and
the static frame of reference was thought to be preferable.
The second difference being the values used for the row in the trans
form matrix, with Ryan, Lorenz and De Doncker[12] using 1/(2 2) and
Zhang[14] using 1/2. The value used in this row determines the range
of possible values that can take, and for this derivation of the 4-leg
matrix converter, the value of 1/(2 2) was chosen because this means that
the 4 primary vectors, a, b, c and n, are all the same length. This is not
the case with the values used by Zhang. While this is not a problem
with the forms of the solution proposed by Zhang for the 4-leg inverter,
it was found that having the primary vectors of the same length was
advantageous for the 4-leg matrix converter.
5.4
Now that the input and output spaces have been initially set out, and
the primary space vectors in each area dened, the next stage in the
derivation of the 4-leg matrix converter is to ascertain the complete set
of switching states that are possible. These switching states are those
which obey the fundamental rules for any matrix converter, that two
input phases cannot be shorted together, and no output leg can be open
52
b+c
c+a
Table 5.1: Rotating Switching States for the 4-Leg Matrix Converter
Rotating
Vectors
B
A
C
B
A
C
B
A
C
B
A
C
B
A
C
B
A
C
C
C
A
A
B
B
C
C
A
A
B
B
C
C
A
A
B
B
A
A
A
A
A
A
B
B
B
B
B
B
C
C
C
C
C
C
54
-V AB
0
VCA
-V AB
0
VCA
0
V AB
-VBC
0
V AB
-VBC
VBC
-VCA
0
VBC
-VCA
0
VCA
VCA
0
0
-V AB
-V AB
-VBC
-VBC
V AB
V AB
0
0
0
0
-VCA
-VCA
VBC
VBC
Input Current
IA
IB
IC
Ia +Ib
Ic
-Ia -Ib -Ic
Ia +Ic
Ib
-Ia -Ib -Ic
Ib +Ic
Ia
-Ia -Ib -Ic
Ic
Ia +Ib -Ia -Ib -Ic
Ib
Ia +Ic -Ia -Ib -Ic
Ia
Ib +Ic -Ia -Ib -Ic
Ia +Ib -Ia -Ib -Ic
Ic
Ia +Ic -Ia -Ib -Ic
Ib
Ib +Ic -Ia -Ib -Ic
Ia
Ic
-Ia -Ib -Ic Ia +Ib
Ib
-Ia -Ib -Ic Ia +Ic
Ia
-Ia -Ib -Ic Ib +Ic
-Ia -Ib -Ic Ia +Ib
Ic
-Ia -Ib -Ic Ia +Ic
Ib
-Ia -Ib -Ic Ib +Ic
Ia
-Ia -Ib -Ic
Ic
Ia +Ib
-Ia -Ib -Ic
Ib
Ia +Ic
-Ia -Ib -Ic
Ia
Ib +Ic
-Ib -Ic
-Ia -Ic
-Ia -Ib
-Ia -Ib
-Ia -Ic
-Ib -Ic
Ia
Ib
Ic
Ic
Ib
Ia
Ia
Ib
Ic
Ic
Ib
Ia
Ib
Ia
Ia
Ib
Ic
Ic
-Ia -Ic
-Ib -Ic
-Ib -Ic
-Ia -Ic
-Ia -Ib
-Ia -Ib
Ib
Ia
Ia
Ib
Ic
Ic
Ic
Ic
Ib
Ia
Ia
Ib
Ic
Ic
Ib
Ia
Ia
Ib
-Ia -Ib
-Ia -Ib
-Ia -Ic
-Ib -Ic
-Ib -Ic
-Ia -Ic
55
Table 5.2: Stationary and Zero Switching States for the 4-Leg Matrix
Converter
Zero
Vectors
IA
0
0
0
Input Current
IB
0
0
0
IC
0
0
0
+1
-1
+2
-2
+3
-3
B
A
C
B
A
C
B
A
C
B
A
C
B
A
C
B
A
C
V8
V7
V8
V7
V8
V7
V AB
-V AB
VBC
-VBC
VCA
-VCA
0
0
0
0
0
0
0
0
0
0
0
0
I6
I3
I2
I5
I4
I1
Ia
-Ia
0
0
-Ia
Ia
-Ia
Ia
Ia
-Ia
0
0
0
0
-Ia
Ia
Ia
-Ia
+4
-4
+5
-5
+6
-6
B
A
C
B
A
C
A
B
B
C
C
A
B
A
C
B
A
C
B
A
C
B
A
C
V4
V11
V4
V11
V4
V11
0
0
0
0
0
0
V AB
-V AB
VBC
-VBC
VCA
-VCA
0
0
0
0
0
0
I6
I3
I2
I5
I4
I1
Ib
-Ib
0
0
-Ib
Ib
-Ib
Ib
Ib
-Ib
0
0
0
0
-Ib
Ib
Ib
-Ib
+7
-7
+8
-8
+9
-9
Stationary
Vectors
A
B
B
C
C
A
B
A
C
B
A
C
B
A
C
B
A
C
A
B
B
C
C
A
B
A
C
B
A
C
V2
V13
V2
V13
V2
V13
0
0
0
0
0
0
0
0
0
0
0
0
V AB
-V AB
VBC
-VBC
VCA
-VCA
I6
I3
I2
I5
I4
I1
Ic
-Ic
0
0
-Ic
Ic
-Ic
Ic
Ic
-Ic
0
0
0
0
-Ic
Ic
Ic
-Ic
+10
-10
+11
-11
+12
-12
A
B
B
C
C
A
A
B
B
C
C
A
B
A
C
B
A
C
B
A
C
B
A
C
V12
V3
V12
V3
V12
V3
V AB
-V AB
VBC
-VBC
VCA
-VCA
V AB
-V AB
VBC
-VBC
VCA
-VCA
0
0
0
0
0
0
I6
I3
I2
I5
I4
I1
Ia +Ib
-Ia -Ib
0
0
-Ia -Ib
Ia +Ib
-Ia -Ib
Ia +Ib
Ia +Ib
-Ia -Ib
0
0
0
0
-Ia -Ib
Ia +Ib
Ia +Ib
-Ia -Ib
+13
-13
+14
-14
+15
-15
B
A
C
B
A
C
A
B
B
C
C
A
A
B
B
C
C
A
B
A
C
B
A
C
V6
V9
V6
V9
V6
V9
0
0
0
0
0
0
V AB
-V AB
VBC
-VBC
VCA
-VCA
V AB
-V AB
VBC
-VBC
VCA
-VCA
I6
I3
I2
I5
I4
I1
Ib +Ic
-Ib -Ic
0
0
-Ib -Ic
Ib +Ic
-Ib -Ic
Ib +Ic
Ib +Ic
-Ib -Ic
0
0
0
0
-Ib -Ic
Ib +Ic
Ib +Ic
-Ib -Ic
+16
-16
+17
-17
+18
-18
A
B
B
C
C
A
B
A
C
B
A
C
A
B
B
C
C
A
B
A
C
B
A
C
V10
V5
V10
V5
V10
V5
V AB
-V AB
VBC
-VBC
VCA
-VCA
0
0
0
0
0
0
V AB
-V AB
VBC
-VBC
VCA
-VCA
I6
I3
I2
I5
I4
I1
Ia +Ic
-Ia -Ic
0
0
-Ia -Ic
Ia +Ic
-Ia -Ic
Ia +Ic
Ia +Ic
-Ia -Ic
0
0
0
0
-Ia -Ic
Ia +Ic
Ia +Ic
-Ia -Ic
+19
-19
+20
-20
+21
-21
A
B
B
C
C
A
A
B
B
C
C
A
A
B
B
C
C
A
B
A
C
B
A
C
V14
V1
V14
V1
V14
V1
V AB
-V AB
VBC
-VBC
VCA
-VCA
V AB
-V AB
VBC
-VBC
VCA
-VCA
V AB
-V AB
VBC
-VBC
VCA
-VCA
I6
I3
I2
I5
I4
I1
Ia +Ib +Ic
-Ia -Ib -Ic
0
0
-Ia -Ib -Ic
Ia +Ib +Ic
0
0
-Ia -Ib -Ic
Ia +Ib +Ic
Ia +Ib +Ic
-Ia -Ib -Ic
56
5.4.1
Now, each of the switching states within Table 5.2 denes a particular
space vector within the space, and are shown plotted in Figure 5.5.
So, if switching state +1 from Table 5.2 is examined, it can be seen that input phase A is switched to output leg a, while input phase B is switched
to the remaining output legs bcn. Looking at the output voltages it is
easy to see that this gives
Va = VA
Vb = VB
Vc = VB
Vn = VB
Transforming this into the space using (5.3.4) gives
57
(5.4.1)
1
2
Vo = 0
3
2 2
1
2
3
2
1
2 2
VA
0
1
V
B
2
= VAB 0
0
3
VB
3
1
2 2
2 2
VB
1
2
23
1
2 2
(5.4.2)
This same result can also be achieved using the primary voltage vectors
as set out in equation (5.3.6), where a is
a = VA 0
3
(5.4.3)
2 2
b + c + n = 2 VB
3
1
2
3
2
1
2 2
2
+ 3 VB
1
2
3
2
1
2 2
0
1
+ 3 VB 0 = 2 VB 0
3
2 2
2 2
(5.4.4)
As can be seen, this is a vector in the opposite direction to a with magnitude VB . The resultant space vector for this switching state is the sum of
the two vectors which is
2
2
3 VA 0 + 3 VB
2 2
1
0
1
22
2
= 3 VAB 0
(5.4.5)
2 2
58
a=8
b+c+n = 4+2+1 = 7
(5.4.6)
59
V14
1/sqrt2
V6
V12
V4
2/3sqrt2
V10
1/3sqrt2
V2
V8
0
V7
V13
V5
-1/3sqrt2
V11
V3
-2/3sqrt2
V9
-1/sqrt2
V1
5.4.2
(5.4.7)
2
I6 = Ia
3
1
0
2
Ia
3
1
2
3
2
I6 = Ia
(5.4.8)
1
3
2
which denes a space vector with magnitude Ia and a xed phase
3
displacement of . If this process is repeated for each of the switching
6
states, as seen in the right-hand columns in Table 5.2, it can be seen that
the input current only has six space vectors, labelled I1 to I6, and that
each can be generated using any of seven switching states. These vectors
can be seen plotted in the space in Figure 5.7, and it can be seen that
although the hexagonal shape made by the vectors is the same, with an
angle of between each vector, the overall space has been rotated by
3
6
to that seen in Figure 5.4.
I2
I3
Bi
I1
Ai
I4
I6
Ci
I5
Figure 5.7: Stationary Current vectors plotted in space
As with the voltage vectors above, this means that there will always be a
set of switching states that can be used to generate any particular input
current space vector irrespective of the required output space vector. It
is this ability to decouple the input vectors from the output vectors that
61
5.5
With all the input and output vectors now dened in terms of the available switching states, the next stage is to identify which of these vectors
is required generate the demanded output voltage. This is initially done
by calculating which input and output sectors the input current and
output voltage vectors reside within.
Looking at the output rst, which as previously stated has four output
legs and is transformed using equation (5.3.4) into a three-dimensional
() space as shown in Figure 5.6. From Table 5.2, and as shown in Section 5.4.1, there are fourteen xed space vectors describing this volume[11,
12, 14]. On closer inspection it can be seen that this volume is actually
a superset of the two-dimensional plane, which can be seen in Figure 5.8 where the space is projected on the plane[11] by looking
along the axis.
63
Va (120)
Vb (164)
Vd =
Vc (44)
Vn (0)
(5.5.1)
X (120)
X (120)
X (0)
= 0
3
2 2
1
2
3
2
1
2 2
Va (120)
V (164)
b
Vc (44)
Vn (0)
1
2
3
2
2 2
(5.5.2)
As has been stated previously in this section, and shown in Figure 5.8,
by looking along the -axis this three-dimensional space can be seen
projected on the plane. This projection is in the same form as the input
voltage plane, and as such can be seen to be made up of six sectors. By
comparing (5.1.5) and (5.3.4), and ignoring the third dimension in (5.3.4),
the transforms are then identical. This allows the output voltage vector
to be easily converted into the 2-dimensional plane by ignoring the
axis, and from here it is trivial to calculate the phase angle in this plane
by using equation (5.5.3) below, which species which prism the output
voltage vector lies within.
1,
2,
3,
Prismo =
4,
5,
6,
for 0 <
for
for
3
< 2
3
3
2
3
<
4
3
< 5
3
for <
for
for
4
3
5
3
(5.5.3)
< 0
64
= arctan
X
X
= arctan
140
140
=
4
4
(5.5.4)
Taking this result and comparing it with (5.5.3), it can be seen that this
output demand voltage lies within Prism 6.
Once the prism has been determined, the exact tetrahedron containing
the demand vector needs to be found. This is achieved by using a method
rst described by Zhang[14], where the tetrahedron is found using a
lookup table, and the one required for the 4-leg matrix converter is shown
in Table 5.3. Along with specifying which tetrahedron the demand vector lies within, this table also gives the three output space vectors which
make up the vertices of that tetrahedron, and these are the space vectors
required to be able to create any demand vector within that tetrahedron.
Table 5.3: Tetrahedron Lookup Table
Tetrahedron
1
3
Prism
4
Vectors
V1
V9
V13
Vectors
V1
V5
V13
Vectors
V1
V5
V7
Vectors
V1
V3
V7
Vectors
V1
V3
V11
Vectors
V1
V9
V11
V V V
an bn cn
V V V
an bn cn
V V V
an bn cn
V V V
an bn cn
V V V
an bn cn
V V V
an bn cn
2
Vectors
V8
V9
V13
Vectors
V4
V5
V13
Vectors
V4
V5
V7
Vectors
V2
V3
V7
Vectors
V2
V3
V11
Vectors
V8
V9
V11
V V V
an bn cn
+
V V V
an bn cn
+
V V V
an bn cn
V V V
an bn cn
V V V
an bn cn
V V V
an bn cn
Vectors
V8
V12
V13
Vectors
V4
V12
V13
Vectors
V4
V6
V7
Vectors
V2
V6
V7
Vectors
V2
V10
V11
Vectors
V8
V10
V11
V V V
an bn cn
+ +
V V V
an bn cn
+ +
V V V
an bn cn
+ +
V V V
an bn cn
+ +
V V V
an bn cn
+ +
V V V
an bn cn
+ +
Vectors
V8
V12
V14
Vectors
V4
V12
V14
Vectors
V4
V6
V14
Vectors
V2
V6
V14
Vectors
V2
V10
V14
Vectors
V8
V10
V14
V V V
an bn cn
+ + +
V V V
an bn cn
+ + +
V V V
an bn cn
+ + +
V V V
an bn cn
+ + +
V V V
an bn cn
+ + +
V V V
an bn cn
+ + +
Zhang found that if you know the prism that the demand vector lies
within, then it is possible to nd the correct tetrahedron by matching the
signs of the three line-neutral voltages Van , Vbn , Vcn for the demand with
the polarities of each of the switching vectors which make up the vertices
of the four tetrahedrons within a single prism. So, from the example
above, where it has already been shown that the output demand voltage
is within Prism 6, the phase-neutral demand voltages are
65
V
an
Vbn
Vcn
120
= 164
44
(5.5.5)
Now if these are compared with the signs of the tetrahedrons which go
to make up Prism 6 in Table 5.3, this gives the result that the output
demand voltage is contained within tetrahedron 3 of that prism, and the
table then gives the three bounding space vectors as being V8, V10 and
V11.
5.6
Moving onto the input current, and once again looking at Figure 5.7 it
can be seen that one area is shaded, and this shows the smallest bounded
area for the input current space. Any space vector within this area can
be generated by combining the two stationary space vectors that bound
either side, and this can be seen in Figure 5.10.
As before, these bounded areas are called sectors and are numbered as
shown in Figure 5.10.
For the matrix converter where the primary concern is controlling the
output voltage, while also controlling ensuring that the converter has a
unity input displacement factor, it is then not the input current magnitude that is directly controlled, as this is dependant on both the output
voltage and the characteristics of the load, instead it is the phase of the
input current which needs be controlled for a unity displacement factor.
Hence the input current vector will be maintained in phase with the
input voltage vector. This phase relationship allows a simple determination of which of the input current sectors the demand vector lies within
for any set of input voltages[4], as it will have the same phase angle as
the input voltage vector, and where the input current sectors are dened
as
66
3
I1
I3
VA
VB
VC
i I5
i I6
ii
i
1
I4
Vi
I6
I5
Figure 5.10: Input Current Plane showing input current vector ii
1,
2,
3,
Sector =
4,
5,
6,
for
for
for
for
for
for
11
6 < 6
6 < 2
5
2 < 6
5
7
6 < 6
7
3
6 < 2
3
11
2 < 6
(5.6.1)
with being the phase angle of the input voltage vector within the
plane. This can be seen in Figure 5.10 where the input voltage phasor
is within the shaded area, which is dened as current sector 6 and the
required input space vectors are I5 and I6.
67
5.7
Now that both the input and output sectors are known for the required
output voltage and input current phase, these are then used to dene
the set of switching states that are needed. However, due to the extra
output leg this is more complex than with the 3x3 matrix converter.
The switching states are found by comparing the two sets of possible
switching states that can be used to generate the input current vector
and output voltage vector, and by using the measured input line-to-line
voltages to select the correct set.
For the output, as has been described above in Section 5.4.1, the entire
output voltage space is dened by fourteen vectors and using the symmetry between the vectors themselves and the switching states used to
create them, it is easy to see that for each output space vector there are
42
three ( 14 ) associated switching state pairs, so as there are three output
space vectors, this then gives nine possible switching state pairs that can
be used within any one tetrahedron.
In a similar manner for the input, the input space is dened by six space
vectors and using the inherent symmetry between the vectors themselves, and switching states, as set out in Sections 5.4.1 and 5.4.2, it
can be seen that for each input current space vector there are seven
( 42 ) associated switching state pairs. So, as there are two input current
6
space vectors for each input current sector, this gives fourteen possible
switching state pairs that can be used for any one stationary space vector.
Continuing to the examples from above, where the input current is in
sector 6, so requiring input vectors I5 and I6 and the output voltage in
tetrahedron 3 of prism 6, requiring output vectors V8, V10 and V11.
Then, referring to Table 5.2 it is possible to select all of the possible
switching states that are able to generate the required vectors, remembering that the vectors themselves come in positive/negative pairs and
so the switching states associated with both of these should be used here.
So for the example above the Table 5.4 shows all the possible switching
states that can be used to generate the different input and output vectors. However, only those switching states which correspond to both the
input current and output voltage vectors can possibly be used with this
combination of input current sector and output tetrahedron, and these
are shown as the Matching States in Table 5.4.
Looking at these six matching switching state pairs in Table 5.2, where
the required output vectors, V8, V10 and V11, can be created by these
68
Output
V8
Current Sector 1
Voltage Sector 6:3
V10
V11
V8 :
1, 2, 3
I5 : 2, 5, 8, 11, 14, 17, 20
V10 : 16, 17, 18
I6 : 1, 4, 7, 10, 13, 16, 19
V11 :
4, 5, 6
1, 2, 4, 5, 16, 17
+1, 2, 4, +5, +16, 17
I1
I6
matching switching states, however the value of the switched line-toline voltage will dictate which of these 12 switching states are actually
used. Once again using the example above, it is known that the input
current vector is within sector 6, where as can be seen in Figure 5.11, VAB
will always be positive and VBC will always be negative. Looking further
at Figure 5.11 it can be seen that for every input current sector, there is
always a single line voltage which is always positive, and is always the
most positive line voltage. There is also a single line voltage which is also
always negative, and this is always the most negative line voltage. It is
the point where the third line voltage either become the most positive,
or the most negative, when the input current sector changes.
1
1
VAB
VBC
VCA
VA
VB
VC
5
6
7
6
3
2
11
6
Figure 5.11: Line and Phase Voltage plots showing current sector
Now looking at the switching state pair 2 that will be used to generate
69
5.8
vd
o
v3
o
o
v2
Figure 5.12: Demand Vector generation from the 3 Space Vectors
From the vector diagram given in Figure 5.12 the following equation can
be written down for the demand voltage
vd = o + o + o
70
(5.8.1)
V1
V2
V3
V4
V5
V6
V7
c+n
b+n
b+c
b+c+n
nnpp
npnn
npnp
nppn
nppp
1
-3
1
-3
-1
3
-1
3
2
-3
-2
3
1
- 3
1
- 3
1
-
3 2
3 2
3
2
3 2
3
1
3 2
3 2
V8
V9
V10
V11
V12
V13
V14
V15
3 2
pnpp
ppnn
ppnp
pppn
pppp
2
3
2
3
1
3
1
3
1
3
1
3
1
- 3
1
- 3
3 2
3 2
3 2
3
1
3 2
3
2
3 2
3 2
o = vo v1
(5.8.2)
o = vo v2
(5.8.3)
o = vo v3
(5.8.4)
where vo , vo and vo are the magnitudes, and v1 , v2 and v3 are the base
vectors in the directions of the three output space vectors. These base
vectors are the direction vectors for the 14 stationary output switching
vectors, along with the 2 zero state vectors, and these are all generated
from the primary vectors (5.3.6) as shown in Section 5.4.1 and listed in
Table 5.5.
So, it is known that the smallest bound volume in the output space is the
tetrahedron, and that it requires 3 stationary voltage vectors to dene the
vertices of this tetrahedron. From Section 5.7 it is known that for each of
the 3 stationary vectors are themselves created by using two switching
states. This therefore means that for each of the vectors o , o and o ,
there are two switching states associated with them, so therefore
71
I
o = vo v1 = vo v1 I + vV I v1 V I
o
o =
o =
(5.8.5)
I
v o v2 = v o I v2 I I + vV v2 V
o
I
IV
vo v3 = vo I I v3 I I I + vo v3 IV
(5.8.6)
(5.8.7)
I
where I to V I are the duty cycles for the six switching states, and vo to
vV I are the instantaneous voltages generated at the output legs.
o
i1
vi
III
III i i
I I +
ii
IV
ii I V
II
II i i
ii
+V
ii V
+V
I V
ii I
i2
Figure 5.13: Input Current Vector generation from the 2 Space Vectors
For in the input current vector, the situation is slightly different. As
stated above, the converter is not directly controlling the magnitude of
the input current vector, instead it is controlling the phase. As this is
done to give the converter a unity displacement factor this means that
the phase angle of the input current should match the phase angle of
the input voltage, . So, as the input current vector is required to
follow the input voltage vector and looking at Figure 5.13, the following
equations, (5.8.8) to (5.8.10), can be written down. These equations force
a zero value onto the perpendicular component of the required current
vector with respect to the input voltage vector. This perpendicular component is signied by the je ji term. This is done in the same way as for
the 3x3 matrix converter[4] and so gives
72
V
iiI I + ii I V I je j i e j(Ki 1) 3 = 0
V
iiI I I I + ii V je j i e j(Ki 1) 3 = 0
(5.8.8)
(5.8.9)
(5.8.10)
where Ki is the input current sector and i is the input current phase
angle measured with respect to a line bisecting the input current sector,
and is limited by
i
6
6
(5.8.11)
|iiI | I cos i
3
= |iiV I |V I cos i +
3
(5.8.12)
then due to the current stiff nature of the output of the matrix converter,
and that the two switching states being used to create a single output
voltage space always switch the same current, as shown in Table 5.2,
this means that
|iiI | = |iiV I |
therefore
73
(5.8.13)
cos i +
I
=
V I
cos i
I I I
I I
= IV
V
(5.8.14)
I
vo I
cos i +
cos
+ vV I I .
o
(5.8.15)
5.9
The nal stage of this derivation is to determine the required duty ratios
for each of the 6 switching states.
Initially, the length of each of the three vectors o , o and o which
produce the demand voltage is required. One solution to this problem is
given by Zhang[14] by using a lookup table, however, while this works
well for the 4-leg matrix converter, it was not thought to be general
enough for use with the 4-leg matrix converter. So, a more general approach to solving this problem was sought, initially using trigonometry
in a similar fashion to that employed by Casadei[4] for the 3x3 matrix
converter. This initial method was unsuccessful as, while it was possible
to perform the calculation, the addition of the third dimension complicated things greatly, and on a practical level the approach did not appear
to be simple, or suitable, for implementation in a DSP.
After further investigation a solution was found using the vector nota74
v o v1
vd
v3
O
v2
(5.9.1)
where P0 is a point in the plane and n23 is the normal to the plane, and is
dened as
n23 = v2 v3
(5.9.2)
In this case, and that of every plane that is of interest, the plane always
passes through the Origin, so the equation then becomes
n23 ( P) = 0
(5.9.3)
(5.9.4)
where P1 and P2 are two points on the line and ( P2 P1 ) denes the
direction, and so for every value of u there is a point P that lays on
the line. In this case we want to know the distance, in the direction of
vector v1 , from the tip of the demand vector vd to the plane formed by
75
P1 = vd
P2 P1 = v1
therefore
P = v d v o v1
(5.9.5)
(5.9.6)
n23 vd
n23 v1
(5.9.7)
(5.9.8)
where is the angle between the two vectors a and b. So, expanding
equation (5.9.7) now gives
vo =
(5.9.9)
where 1 is the angle between the vector v1 and the normal vector n23 ,
and o is the angle between the demand vector vd and the normal vector
n23 .
As vector v1 is a xed space vector, it is therefore xed in relation to
the plane Ov2 v3 and likewise, it is xed in relation to the normal of this
plane. Therefore |v1 | cos 1 is a constant when working in this tetrahedron, and it can easily be shown that, due to the symmetry between the
space vectors, it is in fact constant for all tetrahedra within this space,
and for all stationary space vectors, and it is equal to
1
|v x | cos x =
3
76
(5.9.10)
3|vd | cos o
(5.9.11)
I
3|vd | cos o = vo I + vV I I .
o
cos i
cos i +
(5.9.12)
3| v d |
cos o cos i +
3
I
vo cos i + + vV I cos i
o
3
(5.9.13)
Now, once again using the example from above, where the three output
space vectors being used are V8 , V10 and V11, if it is taken that v1
corresponds to V8 then the two switching states that are being used to
generate this vector are +1 and 2 as shown in Table 5.2.
For switching state 2, this produces input space vector I5, and as this
would correspond to the lower of the two vectors when looking at the
sector as in Figure 5.13, referring back to (5.9.13) it can be seen that
the switching state is therefore controlled by duty cycle V I . When this
switching state is selected, the input phases VB and VC are switched onto
the output, so the voltage, vV I , generated across the output legs which is
o
dened by
VBC = VB VC
(5.9.14)
2
3
Vi cos i
4
3
(5.9.15)
where i is the input voltage phase angle. Then using the cosine sum-toproduct identity gives
77
. sin
3
= 3.Vi sin i
3
5
= 3.Vi cos i
6
4
3
(5.9.16)
However as the equations which dene the input current duty cycles
(5.8.8) to (5.8.10) use a phase angle related to a line which bisects the
current sector, the input voltage phase angle i needs to be referred to
the same line, therefore
i = i
(Sectori 1)
3
(5.9.17)
where Sectori is in the input current sector, and so as the input voltage
vector is within sector 6 equation (5.9.16) becomes
VBC
5 5
3
6
= 3.Vi cos i +
3
VBC =
3.Vi cos i +
(5.9.18)
In the example the input voltage vector is sat within input current sector
6, and as can be seen in Figure 5.11, this means that the voltage VBC is
negative. However switching state 2 is being used and this switches
VB C onto the output, so
(5.9.19)
3.Vi cos i
(5.9.20)
78
3.Vi cos i +
(5.9.21)
I =
|vd |
Vi cos i +
cos o cos i +
3
cos i + + cos i
3
cos i
(5.9.22)
cos i +
+ cos i
cos i
6
3
2
3
1
= cos i + i +
+ cos i i
2
2
6
+ cos i + i
+ cos i i +
(5.9.23)
2
6
cos i +
cos i + i +
+ cos i + i
2
2
2
+ cos i i +
+ cos i i
6
6
(5.9.24)
Using the identity cos ( + /2) = cos ( ), and then cancelling gives
cos i + i +
2
+ cos i i
6
1
cos i i
=
2
cos i + i +
2
+ cos i i +
6
+ cos i i +
6
(5.9.25)
This leaves two cosine terms, both using i i which is the dened as ,
the phase lag between the input voltage and input current. Substituting
79
cos
+ cos +
2
6
6
1
=
2 cos () cos
2
6
3
=
cos ()
2
(5.9.26)
(5.9.27)
In a similar way the equations for the other ve duty cycles can be calculated as
II
I I I =
IV =
V =
V I =
cos
3 Vi
2 |v | cos o cos( i + )
3
d
Vi
cos
3
2 |vd | cos o cos( i )
3
cos
3 Vi
2 |v | cos o cos( i )
3
d
Vi
cos
3
2 |vd | cos o cos( i )
3
cos
3 Vi
(5.9.28)
(5.9.29)
(5.9.30)
(5.9.31)
(5.9.32)
It should be noted that |vd | in these equations is the the length of the
demand vector in the space, and is not the peak voltage of the
demand. However, if the demand voltages do form a balanced 3-phase
set then the length of |vd | equates to the peak phase voltage.
5.10
VA
300
VB
VC
200
Vouta
100
Voutb
Voutc
0
-100
-200
-300
-400
0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
time(s)
Figure 5.15: Input Phase Voltage waveforms with the Output Voltage
Waveforms contained with the envelope
So, if a general, balanced or unbalanced, set of voltages is dened such
that
Va = Va sin ()
2
3
4
Vc = Vc sin
3
Vb = Vb sin
(5.10.1)
the difference between these phases are the respective line voltages, which
81
Vab = Va Vb
Vbc = Vb Vc
Vca = Vc Va
2
3
2
4
= Vb sin
Vc sin
3
3
4
= Vc sin
Va sin ()
3
= Va sin () Vb sin
(5.10.2)
(5.10.3)
Now, taking the exponential form of sin such that
sin ( ) =
e j e j
2j
(5.10.4)
and substituting this into the equation for Vab in (5.10.2) above gives
2
2
e j e j
e j( 3 ) e j( 3 )
Vab = Va .
Vb .
2j
2j
(5.10.5)
(5.10.6)
and using the exponential form of sin once more from (5.10.4)
Vab = Va Vb .e j
2
3
. sin
(5.10.7)
Now, as
e j = cos + j sin
and solving for =
2
3
Vab =
(5.10.8)
jVbi .
3
2
. sin
(5.10.9)
82
|Vab | =
Va +
Vb
2
+ Vb .
sqrt3
2
(5.10.10)
and
ab = arctan
Vb .
sqrt3
2
Vb
2
Va +
(5.10.11)
|Vab | =
(5.10.12)
The same exercise can be performed for each of the other two line-line
voltages giving
|Vbc | =
|Vca | =
3Vx
(5.10.14)
Looking at the input voltages rst, and plotting the resultant line-line
voltages in Figure 5.16, it can be seen that the difference between the
most positive and most negative input phase voltages is the envelope of
the rectied input voltage waveforms.
The entire area under the rectied waveform could be used to generate the output voltage, however, in order to guarantee that it is always
possible to produce the desired output, the output line-line voltage must
always be less than the minimum line-line voltage. This is because at any
instant in the input voltage cycle the converter must be able to generate
any point in the output voltage cycle, so the peak output line voltage
83
VAB
VBC
VCA
5
6
7
6
3
2
11
6
3Vi sin
3
3
3Vi
2
3
Vi
2
Vo(line)
(5.10.15)
and as
Vo(line) =
3Vo( phase)
(5.10.16)
Combining these two equations gives the well known voltage transfer
function for space vector modulation
84
3
2
(5.10.17)
Exactly the same procedure can be used with an unbalanced set of output voltages as well, however in this case there will be a different magnitudes for each of the output line voltages. This will therefore mean
that there are three different results for the voltage transfer ratio, and so
some care must be taken to ensure that the correct ratio is chosen.
If an unbalanced set of output voltages are demanded such that
Vo
3
2Vo
Vb =
3
Vc = Vo
Va =
(5.10.18)
with Vo being the largest of the three output voltage values. Then using
the equations laid out in (5.10.12) and (5.10.13) the following results are
obtained
|Vab | =
Vo
3
|Vbc | =
2Vo
3
|Vca | =
Vo2
Vo 2Vo
+ .
+
3 3
2
2Vo
.Vo + Vo2
3
Vo
+ Vo . +
3
Vo
3
7Vo
9
19Vo
9
13Vo
9
2Vo
3
7
Vo
3
19
=
Vo
3
(5.10.19)
13
=
Vo
3
and combining these results with the result in equation 5.10.16 then gives
3 3
2 7
3 3
2 19
3 3
2 13
q ab
1.70
qbc
1.03
qca
1.25
85
(5.10.20)
(5.10.21)
86
(5.10.22)
2 v 1
d
cos o cos( i + ) + cos( i
3
3 Vi cos
) +
3
) +
3
) 1
3
(5.10.23)
Given that
cos i +
+ cos i
3
3
= cos i
(5.10.24)
3 cos
1
2 cos i (cos o + cos o + cos o )
(5.10.25)
3
cos
2
(5.10.26)
87
5.11
At the time when this work started, the four-leg matrix converter was a
novel circuit for which there was no available prior work, and as such
the work here is entirely based on the extension of the methods used
for modulation in the 3x3 matrix converter and the four-leg inverter.
However, after the rst publication of the basic derivation [32] there have
since been two other approaches to the problem of modulating the fourleg matrix, both of which have originated from the Power Electronic and
Machine Control (PEMC) group at the University of Nottingham.
The rst approach is an extension of the Alesina/Venturini optimum
method [30], as it was perceived that the SVM based method described
within the original work was too computationally intensive to be able to
implement with any reasonably fast switching frequency. While this was
correct with the equations given in the original paper [32], the simplication to the calculation of the duty cycles described in this chapter, and
the further simplications which will be described in Chapter 6 mean
that it is believed that this is no longer the case.
With the difference in computational effort between the extended Alesina
/Venturini optimum method and the method described in this work
now being negligible, there is thought to be very little difference in performance between them, with the choice of which one to use being mostly
down to personal preference.
The second, and possibly most interesting, of the alternative modulation
methods is the one developed by Yue Fan [34, 35]. For this method it
separates the two functional halves of the matrix converter, the rectier
and inverter, and deals with each separately, joining them by a ctitious
DC link. Figure 5.17 below shows a diagram of a 4-leg matrix converter
using this technique, and as can be seen there is a signicant difference to
the circuit layout for this type of matrix converter as it looks a lot closer
to a normal rectier/inverter circuit. However, as can be seen, there is
no DC link capacitor and so this is a type of direct converter just as the
more standard 4-leg matrix converter used in this work, although due to
the two-stage nature this is sometimes called an indirect converter.
While this looks very much like a standard rectier/inverter setup, the
link between the two is not actually a DC link as it varies depending on
the switching of the rectier stage, and this is where the ctitious DC
link name comes from. The main job of the rectier stage in this circuit
88
Input
Fictitious
DC Link
Output
Figure 5.17: Circuit Diagram of the 4-Leg Matrix Converter using the
Fictitious DC Link
is to ensure that there is a positive voltage across the link, and to draw
unity displacement input current.
The inverter in this circuit works just as a normal inverter does, but has
the complexity of having to operate without a steady DC link voltage,
and so this complicates the operation. However, by knowing the input
voltages, and controlling the rectication stage the voltage across the
link can be known at all times, and so the inverter switching adjusted
accordingly.
This circuit has a number of advantages over the standard four-leg matrix converter, rstly the obvious one is that the number of switches in
this circuit is reduced from the 24 required in the standard converter, to
20 in this one with its related cost savings. The second advantage is
that due to some further advances in inverter and rectier modulation []
which allow the use of the demand voltages to dene the duty cycles, it
is possible to create a very simple method for calculating the duty cycles.
Apart from the different computational requirements between the three
techniques discussed here, it is thought that, based on their 3-phase
equivalents, all three techniques would give an equivalent output performance.
89
C HAPTER 6
6.1
The initial simulation work was done using the Matlab software package, as this would be a relatively quick and easy way to implement the
equations derived in Chapter 5, allowing them to be checked that they
were valid and that the conclusions were correct.
6.1.1
4-Leg Inverter
The initial stages of the simulation work was taken up by proving that
the chosen simulation method was correct, where the output voltage is
time averaged over a switching period. This was done by simulating
a 3x3 matrix converter in a number of different methods, and checking
that the results for this chosen method were consistent. As the 3x3 matrix converter has been extensively simulated the results for this work
will not be included here, and so the rst simulation covered here is
an implementation of a 4-leg inverter. This would allow the basics of
the 3-dimensional SVM method as dened in Section 2.1 above to be
tested. Checking this would be especially important as it would ensure
that the code which is used for selecting the required output tetrahedron
is correct, and that the method of calculating the duty cycles for the space
vectors is also correct. These need to be shown to work correctly are they
are both required for the simulations of the 4-leg matrix converter.
There are several methods which can be used for calculating the duty
cycles for the 4-leg inverter, Zhang[14] uses a simple lookup table approach, however due to the more complex nature of the calculations
required for the 4-leg matrix converter, it was felt that the more general
approach derived in Chapter 5 should ultimately be used here. However, the look-up table approach would also be implemented to allow
the general method to be proved prior to using it within the 4-leg matrix
converter simulation.
The implementation for the simulation using the both of these methods
91
tetra_calc.m
Calculate output
tetrahedron
Lookup Method
General Method
inverter_lookup.m
inverter_calc.m
inverter_lookup_delta.m
inverter_calc_delta.m
inverter_voltage_lookup.m
inverter_voltage_calc.m
Calculate output
voltages
Calculate output
voltages
Plot results
Figure 6.1: Block Diagram for the Matlab simulation for the 4-Leg
Inverter
This being a converter with 4 output legs, instead of the 3 for the 3x3
matrix, it might be expected that the simulation setup of the demand
voltages would require 4 different demand voltages. However, in practice it is either the output line voltage demands, Vab , Vbc and Vca , or the
phase voltage demands, Va , Vb and Vc , that are used. The value of Vn
in these situations is generally not specied as this is not generally of
interest when setting the demands, with its value being dened by the
92
(6.1.1)
such that
Vdemab(t) =
2
)
3
4
)
= 2Vdem cos(2 f dem t
3
Vdembc(t) =
Vdemca(t)
(6.1.2)
where Vdem is the line-line voltage and f dem is the frequency for the desired output. These line-neutral demand voltages are then converted to
being the relevant phase voltages as
(6.1.3)
using
1
(2Vdemab(t) + Vdembc )
3
= Vdema(t) Vdemab(t)
Vdema(t) =
Vdemb(t)
(6.1.4)
(6.1.5)
Vdema(t) =
2
)
3
4
= 2Vdem cos(2 f dem t
)
3
Vdemb(t) =
Vdemc(t)
(6.1.6)
1
V(t)
2
V(t) = 0
3 1
V(t)
2 2
1
2
1
2
3
2
1
2 2
3
2
1
2 2
Vdema(t)
Vdem
b(t)
Vdemc(t)
(6.1.7)
The next step is to identify which sector, or in this case as the sector is
a volume in space, tetrahedron, the output demand vector lies in,
and the method for doing this is as described in Section 2.1. It by rst
calculating the angle that the demand vector would project on the
plane using
= arctan
V
V
(6.1.8)
3
Prismo =
4
if
if
if
if
if
if
0 o <
3
o < 2
3
3
2
3 o <
o < 4
3
4
5
3 o < 3
5
3 o < 2
(6.1.9)
Tetrahedron 4
0
+
+
+
0
0
+
+
+
+
+
+
pnpn
+ 0 +
pppn
nnpn
0 0 +
pnpn
+ 0 +
nnpn
0 0 +
Tetrahedron 3
pnpn
0
+
0
+
nnpn
nnpn
0 0 +
0
0
-
0
+
+
+
pnpp
0 - 0
2
Tetrahedron 2
0
-
0
-
+
0
0
+
pnpp
pnpp
0 - 0
nnpp
- - 0
nnpn
pnpp
0 - 0
nnnp
Tetrahedron 1
nnpp
- - 0
nnnp
[- - -]
0
-
0
0
-
95
pnpn
pnnn
vd
Prism 6
Tetrahedron 3
pnpp
Figure 6.3: Closeup view of a single prism, showing the Demand Vector
9 end
11 i f ( Vdemand ( 3 ) >= 0 )
sign = sign + 1 ;
13 end
This is the point where the more general method that is going to be
used for the calculation of the duty cycles in the 4-leg matrix converter
differs from this method, in that the information that it gathers from
the lookup table does not include the duty cycle matrix. For the more
general method this matrix is calculated on the y during the simulation.
6.1.1.1
The output from the code above gives us the tetrahedron number which
can then be used, alongside the prism number, to look up the required
switching states, and the information used to calculate the duty cycles,
using Table 6.1.
Taking the matrix dened within the table, the duty cycles are then worked
96
3
Prism
3/2 - 3/2
3/2
1
0
1
0
- 3/2 0
3/2
1
3
0 1/2 - 3/2 -1 -1/2
0
1/2
3/2
-1
-1
0
-1
0
3
0
6,14,7
9,6,3
6,3,14
3/2
3/2
3/2
3/2
0
1
0
1
0
3
3/2
3/2
3/2 -1 -1/2
1
0 1/2
- /2
1/2
3/2
-1
0
-1
- 3/2 -1
-3/2
0
14,2,7
3,14,2
3,12,2
3/2
0
3
0
-1/2
1
0
3
0
3/2
-1 -1/2 - 3/2 1
- /2 - 3/2 0 1/2
1/2
-3/2 - 3/2 0
-1
0
-1
- 3/2 -1
2,10,7
12,2,5
2,5,10
3/2
3/2
3/2
-3/2
0
-1/2
1
-3/2
0
- 3 0 -1
0
-1 -1/2 - 3/2 1
0
1/2
1/2
3/2 -1
0
- 3 0
- 3/2 -1
10,4,7
5,10,4
5,13,4
-3/2 - 3/2 0
-1/2 - 3/2 1
-3/2 - 3/2 0
0
-1 1
0
1
/2 - 3/2 0 -1
1/2
3/2
1/2
3/2
-1
- 3/2 0
- 3/2 -1
4,11,7
13,4,1
4,1,11
0 - 3 0
-1/2 - 3/2 1
0 - 3 0
3/2
0 1/2 - 3/2 1 1
0
1
/2
3/2
1/2
3/2 -1
- 3/2 0
-1
0
-1
Tetrahedron 4
I,II,III
1,9,8
3/2
- 3/2 0
3 0
0
-1/2 - 3/2 1
9,3,8
3/2
3/2
0
3/2
3/2
0
-1/2 - 3/2 1
3,12,8
0
3 0
-3/2 - 3/2 0
1
0
1
12,5,8
3/2 0
-3/2
0
- 3 0
1
0
1
5,13,8
-3/2 - 3/2 0
3/2
- 3/2 0
3/2
-1/2
1
13,1,8
0
- 3 0
3/2
3/2 0
1/2
3/2 1
-
I
(t)
II
= 1
(t)
Vdc
I
(tI)I
1 3
2
2
1
2
3
2
23
3
2
V(t)
1 V(t)
V(t)
0
1
(6.1.10)
Now that both the duty cycles and switching states are known, the output voltages at each time point can be calculated using
Vo =
Va(t)
Vb(t)
Vc(t)
Vn(t)
97
(6.1.11)
0A
0C
1
0
1
0
1
0
1
0
0
0
0
0
0
0
Vectors
1
2
3
4
5
6
7
8
9
10
11
12
13
14
1
0
0
1
0
1
0
1
1
0
1
0
1
0
0
1
1
0
0
1
0
1
1
0
0
1
0
1
0
1
0
1
1
0
0
1
0
1
0
1
1
0
0
1
0
1
0
1
1
0
0
1
1
0
0
1
Vdc
-Vdc
0
0
0
0
-Vdc
Vdc
Vdc
-Vdc
0
0
Vdc
-Vdc
0
0
Vdc
-Vdc
0
0
-Vdc
Vdc
Vdc
-Vdc
-Vdc
Vdc
0
0
0
0
0
0
Vdc
-Vdc
-Vdc
Vdc
0
0
-Vdc
Vdc
Vdc
-Vdc
The results from this calculation being saved into the matrix shown below
(6.1.12)
where
(6.1.13)
98
Voutline
Vout
=
Voutbc(1) .. Voutbc(t1) Voutbc(t)
bc(0)
(6.1.14)
where
(6.1.15)
200
Vdema
Vdemb
150
Vdemc
100
Vouta
50
Voutb
Voutc
0
-50
-100
-150
-200
0.5
1.5
2.5
3.5
4.5
time(ms)
Figure 6.4: Output from the 4-Leg Inverter Simulation, showing the
Calculated and Demand Output Voltages
99
Prismo
sign
10
Tetrao
o
0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
Figure 6.5: Output Voltage Vector angle, Tetrahedron and Prism for the
4-Leg Inverter Simulation
The results demonstrated in these gures show that the method used to
implement the 3-dimensional space vector modulation for the 4-leg inverter is correct. The next step is to implement the more general method
of calculating the duty cycles, which will be used with the 4-leg matrix
converter, and ensure that the outputs from both methods match.
6.1.1.2
With the prism and tetrahedron identied, the next step in calculating
the duty cycles via this method is to calculate the length of the space
vectors at the vertices. This is examined in Section 5.9, and while the
derivation of the lengths of the space vectors v1 , v2 and v3 shown is
correct, and this same method is shared with the inverter, it does not
lend itself to straightforward calculation. This means that a modication
is required for the program to be as efcient as possible, and this is
described below.
As the directions of all three output vectors is known, we can dene a
plane which contains any two of them and the origin Ov2 v3 such that
100
1 9 13
1 5 13
157
137
1 3 11
1 9 11
8 9 13
4 5 13
457
237
2 3 11
8 9 11
8 12 13
4 12 13
467
267
2 10 11
8 10 11
n23 ( P) = 0
8 12 14
4 12 14
4 6 14
2 6 14
2 10 14
8 10 14 ]
(6.1.16)
where n23 is the normal vector to the plane Ov2 v3 and P is any point on
the plane.
Then, using the same method as shown in Section 5.9, vo , is then the
length of the vector which intersects both the demand voltage vd and
the plane Ov2 v3 in the direction v1 . This then gives the equation
n23 (vd vo v1 ) = 0
(6.1.17)
(6.1.18)
Using the equation for the dot product (5.9.8) on the left hand side then
gives
vo |n23 ||v1 | cos 1 = n23 vd
(6.1.19)
(6.1.20)
101
3n23 vd
(6.1.21)
(6.1.22)
v2 v3 =
v 2 v 3 v 2 v 3
v 1 v 3 v 1 v 3
v 1 v 2 v 1 v 2
(6.1.23)
Now, for any single tetrahedron, there will be 3 space vectors which
make up its vertices, using the example from Chapter 5 once again these
would be V8, V10 and V11, and these are dened as
2
3
V8 = 0 V10 =
3 2
1
3
13
3 2
V11 =
1
3
1
3
(6.1.24)
3 2
|n xy | =
2
3
(6.1.25)
So, the length of the normal vector to the plane, irrespective of which 2
of the 3 vectors are used to dene it, is always the same length. Looking
at sim: tetrahedron vectors it can be seen that of the 3 vectors, V8 and
11 are the same length
1
sqrt2
, while 10 is longer
2
3
. It can then be
shown that unlike the 6 space vectors which describe the plane in 2dimensional SVM, the 14 space vectors used in 3-dimensional SVM are
not all equidistant from each other, and in fact this must be true for the
result in (6.1.25) to stand. It is in fact the case that there are two different
length groups of vectors, 8 shorter vectors and 6 longer vectors, and that
every tetrahedron is dened by 2 short and 1 long vector, and so the
result shown in (6.1.25) holds true at all times.
So, substituting (6.1.25) into (6.1.28) now gives
102
3 3
vo = n23 vd
2
(6.1.26)
As both the vectors n23 and vd are known, then the value of their dot
product can be easily calculated using
n23 vd = n23 .vd + n23 .vd + n23 .vd
(6.1.27)
3 3
n23 vd
=
vdc 2
(6.1.28)
3 3
n13 vd
=
vdc 2
3 3
n12 vd
=
vdc 2
(6.1.29)
(6.1.30)
Now, with all the duty cycles now dened, all that is required is to
perform a similar set of calculations as for the lookup method, and to
produce the output phase voltage matrix
Voutan(0)
Voutcn(0)
Voutcn(1) .. Voutcn(t1) Voutcn(t)
103
(6.1.31)
(6.1.32)
(6.1.33)
where
(6.1.34)
Vdema
Vdemb
150
Vdemc
100
Vouta
50
Voutb
Voutc
Vouta
-50
Voutb
Voutc
-100
-150
-200
0.5
1.5
2.5
3.5
4.5
time(ms)
(a) Output Phase Voltage for a 4-leg Inverter using the general method for calculating
the duty cycles
300
Vdemab
Vdembc
200
Vdemca
Voutab
100
Voutbc
Voutca
Voutab
Voutbc
-100
Voutca
-200
-300
0.5
1.5
2.5
3.5
4.5
time(ms)
(b) Output Voltage Vector angle, tetrahedron and prism
Figure 6.6: Line and Phase Output Voltages from the 4-Leg Inverter
Simulation, showing the results from the Lookup and General Method
is entirely due to the way that the order that the switching states are
105
tot
I
0.7
I I
0.6
I I I
0.5
0.4
0.3
0.2
0.1
0
0.5
1.5
2.5
3.5
4.5
time(ms)
(a) Output Phase Voltage for a 4-leg Inverter using the general method for calculating
the duty cycles
0.8
tot
I
0.7
I I
0.6
I I I
0.5
0.4
0.3
0.2
0.1
0
0.5
1.5
2.5
3.5
4.5
time(ms)
(b) Duty Cycles for the General method
Figure 6.7: Duty cycle plots for the Lookup and General methods of
4-Leg Inverter Simulation
selected is different between the two methods, and that if these could be
106
(6.1.35)
For the simulation this is done by setting one of the demand for one of
the output phases, Vdema(t) , to have twice the voltage level of the other
2 phases, and then another output phase, Vdemb(t) , to have a twice the
output frequency of the other 2 phases, as shown in equation 6.1.36.
2
Vdemb(t) = 2Vdem cos(4 f dem t
)
3
4
Vdemc(t) = 2Vdem cos(2 f dem t
)
3
(6.1.36)
Unlike the balanced 3-phase set, using these demand voltages gives the
demand vector a complex path within the space, with the value
going both positive and negative, instead of the vector being entirely
within the plane as with the balanced set of demand voltages.
The results from this simulation can be seen in Figure 6.8, which show
that both the lookup and general method for calculating the duty cycles
are capable of producing an output to match the demand voltages.
From these results it is possible to state that the output from the gen107
108
Vdema
80
60
Vdemc
40
V
Vdemb
Vouta
20
Voutb
Voutc
0
-20
-40
-60
-80
-100
0.5
1.5
2.5
3.5
4.5
time(ms)
(a) Output Phase Voltage for a 4-leg Inverter using the lookup method for calculating the
duty cycles
100
Vdema
80
60
Vdemc
40
V
Vdemb
Vouta
20
Voutb
Voutc
0
-20
-40
-60
-80
-100
0.5
1.5
2.5
3.5
4.5
time(ms)
(b) Output Phase Voltage for a 4-leg Inverter using the general method for calculating
the duty cycles
Figure 6.8: Output Phase Voltage for a 4-leg Inverter when following a
complex set of demand voltages
109
6.1.2
With the results from the 4-leg inverter simulations proving that the
method for simulating the 3-dimensional space vector modulation was
correct, along with the general approach to calculating the duty cycles,
the next stage was to pull together the various different parts from the
above simulations to produce a simulation model for the complete 4-leg
matrix converter. However there were a number of steps that needed to
be overcome before this simulation could be made to work.
With the 4-leg inverter, as it only has a single input voltage, it is only
possible to create a switching vector in a single way. However with the
matrix converter, there are numerous converter switching states which
can create any particular switching vector depending on the state of
the input voltages at that time. This is a problem shared with the 3x3
matrix converter, however, with the greater number of output phases it
increases the possible number of combinations, and makes the selection
slightly more complex.
As with the 4-leg inverter, the switching states that would be required at
any one instant would need to be vertices in the output voltage tetrahedron, but on top of that, they also need to be vertices in the input sector
for the input current. Because any output tetrahedron can be associated
with any input sector, and vice versa, then with six input sectors and
twenty four output tetrahedrons this gives a possible of 144 different
combinations. While it would be perfectly possible to use a large 1:1
lookup table to perform this task, this would lead to a quite large and
unwieldy table, and with some experience from using the lookup tables
in the previous simulations, one that would be hard to nd errors within.
As such another method was sought that could make this signicantly
easier to program, and less likely to error.
Early on in the work it had been noticed that there was a certain amount
of symmetry within the input and output spaces, this can also be seen
in with the 3x3 matrix converter, and so these would be a good place to
start looking to simplify things.
As has been shown in Chapter 5 above, for each output voltage tetrahedron there is a set of possible switching states. Using the example in
Table 5.4 above, the tetrahedron 6 : 3 is made up of the space vectors
V8, V10 and V11, and these can be generated by the following switching
states
110
V8
1 2 3
1 2 3
V7
Tetrahedron 3 : 2 = Vectors V5 = 16 17 18
V4
4 5 6
(6.1.38)
As can be seen, the set of possible switching states that can be used to
create the space vectors for the two tetrahedrons opposite to one another
are the same set of switching states. This is possible because, unlike the
4-leg inverter, there are always possible negative input voltages available, and switching a negative voltage in a particular switching state
produces a vector in the opposite direction.
Now moving onto the input sector, as stated in Section 5.7 above, this
selection of switching states can be narrowed depending on which input
current sector is required. As one of the requirements for this matrix
converter is that the input has a unity displacement factor, it is a simple
case of dening the required input current sector as being the sector
which the input voltage vector resides. Continuing the example from
Table 5.4 above, if the input current sector is 6, this is made up of the
space vectors I6 and I5, and these can be generated by the following
switching states
Sector 1 = Vectors
I6
I5
If this is then also performed for each of the other input current sectors,
as for the output voltage tetrahedrons above, it can be seen that once
again there is a symmetry between sectors. So, the opposite input cur111
Sector 3 = Vectors
I3
I2
Once again it can be seen that the possible switching states that can be
used to generate the pair of opposite sectors are the same, and once again
this is due to having both positive and negative voltages being generated
at the outputs. Using these symmetries it is now possible to reduce the
number of combinations of input and output from 144 to 36 such that
Tetrahedrons
3:2
3
and Sectors = 1, 2, 4, 5, 16, 17
6:3
6
(6.1.41)
(6.1.42)
Know that the input current is in sector 6, due to the unity displacement
factor this means that the voltage vector must also in input current sector
6, and to be in that sector it means that
VA
+ve
Sector 6 = VB = -ve
VC
+ve
(6.1.43)
Looking at Table 5.2 at the 6 switching state pairs it can be seen that
for these switching states the converter only switches between the lineline voltages VAB and VBC . So, during this time the only possible input
voltages are
112
Sector 6 =
VAB
VBC
+ve
-ve
(6.1.44)
Knowing that to generate the correct space vectors for the output voltage
tetrahedron the switching states need to generate the vectors V8, V10
and V11, and once again referring back to Table 5.2, it shows that with
a positive VAB and a negative VBC the following switching states are
needed
Sector 6 =
+VAB
VCA
+1 +5 +16
2 4 17
(6.1.45)
Performing this same analysis for the other three tetrahedron/sector pairs
which share this set of 6 switching state pairs then gives the results seen
in Table 6.4.
Table 6.4: Switching state selection example
Current Sector
Current Sector 3
Current Sector 6
Tetrahedron
3:2
6:3
3
Sectori =
4
if
if
if
if
if
if
11
6 i < 6
6 i < 2
5
2 i < 6
5
7
6 i < 6
7
3
6 i < 2
3
11
2 i < 6
(6.1.46)
Then calculate the output voltage tetrahedron in the same way as for the
4-leg inverter, by rst calculating which prism in space the demand
vector resides in by
3
Prismo =
if
if
if
if
if
if
0 o <
3
o < 2
3
3
2
3 o <
o < 4
3
4
o < 5
3
3
5
3 o < 2
(6.1.47)
114
1
2
3
Prism
4
5
6
1
1
5
9
4
8
12
sign =
2 3
2 3
6 7
10 11
3 2
7 6
11 10
4
4
8
12
1
5
9
Next, using the values already found for Sectori and Tetrao , use another
lookup table, shown in Table 6.6 , to nd the 3 space vectors, v1 , v2 and
v3 , which form the vertices of the tetrahedron
The order that both the space vectors and the switching states is held
is important here because of the relationship between them. As can be
seen in Section 5.9 the relationship is that the switching states ssI and
ssVI are associated with space vector v1 , and likewise ss2 and ss5 with
space vector v2 and then ss3 and ss4 with v3 .
With this relationship, it becomes easy to identify the pairs of switching
states related to each vector, but some care needs to be taken when identifying which switching state to associate with with duty cycle. Comparing the equations for the duty cycles I (5.9.27) and V I (5.9.32), both of
which are associated with the same space vector v1 . As can be seen the
only different between the two equations is that one uses ( i + , and
3
( i + duty cycle is associated with the lower of the two space vectors,
3
in this case I5, while the ( i duty cycle is related to the upper space
3
vector (I6).
From Table 5.4, the switching states +1, 4and + 16 are associated with
space vector I6 and so use the ( i + duty cycles, and switching states
3
2, +5and 17 with space vector I1 and ( i . By performing this
3
analysis for each tetrahedron and input current sector, it was possible
115
7
8
9
10
11
12
1:1
4:4
1:2
4:3
1:3
4:2
1:4
4:1
2:1
5:4
2:2
5:3
2:3
5:2
2:4
5:1
3:1
6:4
3:2
6:3
3:3
6:2
3:4
6:1
V2, V6, V7
V2, V3, V7
V1, V3, V7
V4, V5, V7
V4, V6, V7
to build up the following sets of lookup tables, which would not only
work for the Matlab simulation, but would also be used for the Saber
simulations and the build of the actual converter.
So, using the same values found for Sectori and Tetrao , use the lookup
table, shown in Table 6.7 to determine the base set of switching states,
ssI to ssVI , for that input sector/tetrahedron pair.
The next step is to dene which switching state in each of the selected
pairs is going to be required, and which duty cycle that switching state
is associated with. Once again a lookup table is used, which is shown in
Table 6.8.
116
7
8
9
10
11
12
1:1
4:4
1:2
4:3
1:3
4:2
1:4
4:1
2:1
5:4
2:2
5:3
2:3
5:2
2:4
5:1
3:1
6:4
3:2
6:3
3:3
6:2
3:4
6:1
3, 15, 9, 8, 14, 2
2, 14, 8, 7, 13, 1
1, 10, 7, 9, 12, 3
3, 12, 9, 8, 11, 2
2, 11, 8, 7, 10, 1
1, 10, 19, 21, 12, 3 3, 12, 21, 20, 11, 2 2, 11, 20, 19, 10, 1
19, 16, 7, 9, 18, 21 21, 18, 9, 8, 17, 20 20, 17, 8, 7, 16, 19
4, 16, 7, 9, 18, 6
6, 18, 9, 8, 17, 5
5, 17, 8, 7, 16, 4
4, 10, 7, 9, 12, 6
6, 12, 9, 8, 11, 5
5, 11, 8, 7, 10, 4
4, 10, 19, 21, 12, 6 6, 12, 21, 20, 11, 5 5, 11, 20, 19, 10, 4
19, 16, 1, 3, 18, 21 21, 18, 3, 2, 17, 20 20, 17, 2, 1, 16, 19
4, 16, 1, 3, 18, 6
6, 18, 3, 2, 17, 5
5, 17, 2, 1, 16, 4
4, 13, 1, 3, 15, 6
6, 15, 3, 2, 14, 5
5, 14, 2, 1, 13, 4
4, 13, 19, 21, 15, 6 6, 15, 21, 20, 14, 5 5, 14, 20, 19, 13, 4
At this point everything is known that will allow the duty cycles to be
calculated:
Sectori states what input current sector the input voltage vector lies
within
Prismo states which prism in the output voltage space the demand
vector sit in
Tetrao , which is in a particular tetrahedron in the output voltage
space the demand vector sits within
117
Vectors
1, 2 or 3
4, 5 or 6
ssI ssII ssIII + ssIV + ssV + ssVI +ssIII + ssII + ssI ssVI ssV ssIV
1
v1 = v1
1,
+ssI ssII ssIII + ssIV + ssV ssVI +ssIII + ssII ssI + ssVI ssV ssIV
2
v2 = v2
+ssI + ssII ssIII + ssIV ssV ssVI +ssIII ssII ssI + ssVI + ssV ssIV
3
v3 = v3
or 5
+ssI + ssII + ssIII ssIV ssV ssVI ssIII ssII ssI + ssVI + ssV + ssIV
Sectori
Prismo
Space
ssIV ssV ssVI + ssI + ssII + ssIII +ssVI + ssV + ssIV ssIII ssII ssI
1
v1 = v3
2,
ssIV ssV + ssVI ssI + ssII + ssIII ssVI + ssV + ssIV ssIII ssII + ssI
2
v2 = v2
ssIV + ssV + ssVI ssI ssII + ssIII ssVI ssV + ssIV ssIII + ssII + ssI
3
v3 = v1
or 6
4
+ssIV + ssV + ssVI ssI ssII ssIII ssVI ssV ssIV + ssIII + ssII + ssI
2 v cos x cos( i + )
3
x = d
cos
3 vi
(6.1.48)
As previously stated, this converter is designed to have a unity displacement factor, therefore cos is set to 1. Alongside this, it was shown in
(6.1.25) that the length of the normal vector is always constant, no matter
which combination of the 3 space vectors which dene a tetrahedron are
chosen, so the dot product of the normal vector to the plane nyz and
the demand vector (nd ) becomes
nyz vd =
3
vd cos x = nyz vd
2
118
2
v cos x
3 d
(6.1.49)
x =
vi
nyz vd cos( i
)
3
(6.1.50)
As this work on deriving the various equations took place, each part was
written into a separate Matlab m-le function. This allowed each stage
to be easily checked to ensure that it was correct before then using each
of them be used to build up the overall program to simulate the entire
converter. Figure 6.9 shows each of these different functions, and how
they all t together to form the overall simulation. As can be seen in
the block diagram, the simulation actually performs the calculation of
the duty cycles twice, once by using the equations (5.9.27)-(5.9.32) for
derived in Chapter 5, and the second set is to validate the simplication
of the equations which were derived earlier in this section.
6.1.2.1
Calculate output
tetrahedron
current_sector.m
Calculate input
current sector
matrix_switch_lookup.m
Derived Method
Simplified Method
delta_calc_derived.m
delta_calc_new.m
Vout_calc.m
Vout_calc.m
Calculate output
voltages
Calculate output
voltages
Plot results
qmax =
3
cos
2
(6.1.51)
The rst simulation used a demand voltage of 23 Vin peak with an output
frequency of 100Hz, and the results of the output voltage can be seen in
Figure 6.13, and these are as expected, with the output voltage matching
the demand voltage. With this demand voltage, being the calculated
maximum for the input voltage, the plot of the sum of the duty cycles
should reach a peak of 1 at several points during the simulation when the
output voltage is at its maximum, and the available input voltage at its
minimum. However, looking at the plot of the duty cycles in Figure 6.14a
it can be seen that the total of the duty cycles does not actually reach 1,
120
Vdema
200
Vdemb
150
Vdemc
100
V
Voutan
50
Voutbn
Voutcn
0
-50
-100
-150
-200
0
0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
time(s)
Figure 6.10: 4-Leg Matrix Converter output voltages for the original
derived duty cycle equations
Vdema
200
Vdemb
150
Vdemc
100
V
Vouta
50
Voutb
Voutc
0
-50
-100
-150
-200
0
0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
time(s)
Figure 6.11: 4-Leg Matrix Converter output voltages for the simplied
duty cycle equations
it reaches a maximum of approximately 0.96, and so it appears to not
be following the inequality found in equation (5.10.26). The reason for
this discrepancy is found in Figure 6.14b, which is a plot of the voltage
generated at each of the output legs, including the neutral output leg,
and all referenced to the supply neutral. This shows that the output
121
Sectori
i
5
4
3
2
1
0
-1
0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
time(s)
(a) 4-Leg Matrix Converter Simulation Input Current Vector angle and sector
12
Prismo
Tetrao
10
sign
0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
time(s)
(b) 4-Leg Matrix Converter Simulation Output Voltage Tetrahedron and Sector
Vdema
Vdemb
200
Vdemc
Voutan
100
Voutbn
Voutcn
0
-100
-200
-300
0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
time(s)
that the input frequency has been offset by 12 . The output phase-neutral
voltages remain unchanged from those shown in Figure 6.13, but looking
that the individual leg voltages when referenced to the supply neutral,
shown in Figure 6.15b, the difference with the previous simulation is
easy to see, and that there are several points throughout the cycle where
the peak output voltage equals the available input voltage.
So, as can be seen by the results shown above, while limit to the voltage
transfer ratio of 0.866Vin is true in a general case for a balanced 3-phase
output, there are once again specic times when this is not the case.
Looking at all the results from the Matlab simulations, they show that
the assumptions made, and the derived equations appear to be correct,
with the output voltages Va , Vb and Vc being able to be independently
controlled with respect to the neutral leg voltage. Obviously there are
limitations to these simulations in that this is only simulating the output
voltage, and so the input current control has yet to be tested. So, while
this does not fully validate the design, it gives enough condence in
the operation of the circuit at this point to move onto the next stage,
123
tot
0.9
0.8
I I
0.7
I I I
IV
0.6
0.5
V I
0.4
0.3
0.2
0.1
0
0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
time(s)
(a) 4-Leg Matrix Converter Simulation Duty Cycles when operating with q and
synchronised
400
Vin A
VinB
300
VinC
200
Vouta
100
Voutb
Voutc
Voutn
-100
-200
-300
-400
0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
time(s)
(b) 4-Leg Matrix Converter Simulation Input and Output Voltages, referenced to supply
neutral, when operating with q and synchronised
124
tot
0.9
0.8
I I
0.7
I I I
IV
0.6
0.5
V I
0.4
0.3
0.2
0.1
0
0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
time(s)
(a) 4-Leg Matrix Converter Simulation Duty Cycles when operating with q and
unsynchronised
400
Vin A
VinB
300
VinC
200
Vouta
100
Voutb
Voutc
Voutn
-100
-200
-300
-400
0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
time(s)
(b) 4-Leg Matrix Converter Simulation Input and Output Voltages, referenced to supply
neutral, when operating with q and unsynchronised
125
6.2
The Matlab simulations had so far proved that the equations which had
been derived were providing the expected results at that stage. They
show that the output voltage is able to be controlled, and also demonstrated how the internal variables used for selecting the input sector
and output tetrahedron were being correctly selected. At this point the
simulation was moved on to the next stage by using Saber, which is a
simulation software package from Synopsys. This is a powerful circuit
simulation tool which can be used to simulate complex circuits, both
analogue and digital, and this would allow a more detailed simulation
of the 4-leg matrix converter to be performed, this time being able to
take into account the load on the converter. This meant that in addition
to looking at the output voltage it would also allow the input current
waveforms to be investigated as well.
The rst step in the simulation process with Saber is to design the circuit
which will be simulated. Now, Saber is an incredibly powerful tool, and
it is easy to over-complicate a simulation by going into far too much detail in the circuit design, and while this will provide very accurate results
it would also be very slow, and the vast amount of the information which
it would be able to provide would be unnecessary for this investigation.
So, as always with any simulation, there is a trade-off between the speed
which the simulation runs and the amount of detail which the results
provide.
The circuit for the 4-leg matrix converter is made up of 12 bi-directional
switches, the basic layout of which can be seen in Figure 1.6, with an
example switch being shown in Figure 6.16. Each switch is made up
of 2 diodes and 2 IGBTs, with each of the IGBTs requiring a separate
gate drive circuit, and then also each IGBT has a separate gate drive
signal to ensure that current commutation occurs correctly, using one of
the techniques described in Chapter 4. Now, while a gate drive, with
the logic required to commutate the current correctly, might be a simple
circuit to build within Saber, having 24 of these will increase the number
of elements requiring simulation. Even though this would lead to a very
good model of a working converter, the extra information held in the
results would be unnecessary as the interest in these simulations is in
being able to verify that the derivation and operation of the 4-leg matrix
126
Gate Drive
Circuit
Gxy
Four Step
Commutation
Gate Drive
Circuit
Q1
D2
D1
P1
D4
D3
Q2
Gate1
P2
Figure 6.17: Circuit Diagram used for the Bi-Directional Switch block in
the Saber simulations
be discussed.
The control block shown, as shown in Figure 6.19, is an element which
uses Sabers internal scripting language, MAST, to dene its internal
operation, and then dene how that appears on its outputs. The MAST
language is a hardware description language(HDL) which gives a programmatic way of describing both analogue and digital circuits. For this
simulation the control block is being used to calculate the duty cycles
and then output the respective switch control signals for each of the 12
switching blocks in the converter. These output signals will need to
depend on both the input voltage and the demand voltage/frequency.
There are also a number of additional outputs labelled to0 to to7, these
are used to allow the internal variables within the block to be monitored.
The design of the MAST language template for the control block started
with the Matlab m-les used in Section 6.1.2 above, but as it is not possible to directly run the Matlab simulation les within the simulation
package, these had to be modied. Obviously a reasonable amount of
the Matlab les themselves were also concerned with the mechanics of
the simulation itself, and so needed to be removed anyway, this left
behind the core logic required to calculate the switching states and duty
cycles for the converter. This core part of program could then be modied to meet the requirements of the MAST language. While the basic
128
SAa
SAb
SAc
SAn
SBa
SBb
SBc
SBn
SCa
SCb
SCc
SCn
A
B
C
to0
to1
to2
to3
to4
to5
to6
to7
clk
gnd
gnd
Va
Vb
5m
Va_load
Vc
5m
Vb_load
30
Vn
5m
Vc_load
30
5m
Vn_load
30
A
B
C
to0
to1
to2
to3
to4
to5
to6
to7
clk
gnd
130
+1
2
4
Switching State
+5
+16
17
Output Leg
a b c n
A B B B
C B B B
A B A A
C B C C
A B A B
C B C B
so has a voltage developed across it. So, if the converter is used with the
above switching arrangement, the power loss in the devices in output
leg a will be signicantly higher during this period than those in leg b.
It is therefore an advantage to be able to reduce the number of switching
transitions where possible, and especially as in this case as it is as simple
as rearranging the order in which the switching states are used.
So, taking the switching order from in Table 6.9 above, these states can
then be re-arranged in such a way as to minimise the total number of
switching transitions required, and this is shown in Table 6.9.
Table 6.10: Modied switching state order to minimise the number of
switching transitions
+5
17
2
Switching State
+1
+16
4
Output Leg
a b c n
C B C C
C B C B
C B B B
A B B B
A B A B
A B A A
I I
2
I I I
2
IV
2
V
2
V I
2
4
+16
+1
2
17
+5
Ts
2
Ts
2
132
z = 1 1 2 3 4 5 6
(6.2.1)
tz = Ts .z
(6.2.2)
ZC
+5
17
Switching State
2
+1
+16
4
ZA
Output Leg
a b c n
C C C C
C B C C
C B C B
C B B B
A B B B
A B A B
A B A A
A A A A
The second method starts off looking at the switching states in exactly
the same way, however this time it is also noted that the 3rd (+1) and
4th(2) switching states also have 3 of the 4 outputs switched to a sim133
ZC
+5
17
Switching State
2
ZB
+1
+16
4
ZA
Output Leg
a b c n
C C C C
C B C C
C B C B
C B B B
B B B B
A B B B
A B A B
A B A A
A A A A
Examples of the control signals for both of these possible switching sequences are shown in Figure 6.21. Due to all the output legs having the
same number of switching transitions in the Three Zero method, it was
thought that this would give the better harmonic performance, and so it
was chosen to be the initial method used within the simulation, although
both types of switching sequence would be investigated.
The nal step in the design of the MAST template was to nd a way
to interface the code which calculates the duty cycles and switching
states with the circuit shown in Figure 6.18. The problem that needed
to be overcome was this, how to ensure that the control block only performed its calculations once per switching period while still being able to
change the control outputs multiple times during this sampling period
with high accuracy. The rst part of this problem is relatively simple to
solve, by using an internal clock which is set to the converter switching
frequency, a positive going edge on this clock can be used to trigger
the control block into performing the duty cycle and switching state
calculations. This does not help with the switching of the control outputs
134
I
2
z
3
I I
2
I I I IV V
2
2
2
V I
2
z
3
S Aa
S Ab
S Ac
S An
SBa
SBb
SBc
SBn
SCa
SCb
SCc
SCn
ZC +5 17 2 +1 +16 4
ZA
Ts
2
Ts
2
(a) Two Zero Switching Sequence for the 4-Leg Matrix Converter
z
3
I
2
I I
2
I I I
2
z
3
IV V
2
2
V I z
2
3
S Aa
S Ab
S Ac
S An
SBa
SBb
SBc
SBn
SCa
SCb
SCc
SCn
ZC +5 17 2 ZB +1 +16 4 ZA
Ts
2
Ts
2
(b) Three Zero Switching Sequence for the 4-Leg Matrix Converter
z
3
(6.2.3)
Once this timer triggers, the new set values for this switching state are
loaded onto the outputs and the time to the next switching transition
loaded into the clock.
This method has the advantage that the simulation does not need to
perform any calculations within the control block at any time other than
at one of the clock transitions, to either calculate a new set of duty cycles,
or update the output states, and so this will speed up the simulation
considerably. This nal step completes the MAST template, and a block
diagram showing the functions within it is shown in Figure 6.22.
Having already proved that the duty cycle calculations and switching
state selection derived in Chapter5 are correct for controlling the converter output voltage, by having simulated and tested them using Matlab in Section 6.1.2, the rst step in the Saber simulations was to validate
the MAST control block template against these known results. This was
done using a series of tests to check on the state of the control blocks
internal variables.
6.2.1
The initial testing was used to make sure that the variables representing the input voltage vector were correct, and matched those results
found in the Matlab simulations. Figure 6.24 shows the results from this
136
Figure 6.22: Block Diagram of the MAST template for the Control Block
test where it can be seen that as the input voltage vector angle beta_in
changes, the input current sector isector changes with it in accordance
with equation (5.6.1). The changes in both of these are then also matched
by how the variable beta_in_bar varies, and so by comparing these results to those from the Matlab simulation in Figure 6.12a it can be seen
that they match perfectly.
The output variables representing the demand vector were also checked
along similar lines, with the results being shown in Figure 6.25. Once
again, by comparing the results from this simulation to those from the
Matlab simulation, shown in Figure 6.12b it can be seen that the internal
variables are working as expected, and match the values from the Matlab
simulation. From the results it can be seen that, although the angle that
the output vector makes on the plane is not plotted, the value can be
137
(6.2.4)
for a 3-phase set the value of sign will therefore only be 2 or 3, which
matches the result in Figure 6.25. This, so far, proves that the operation of
the control block MAST template is working in the same way in relation
to identifying the input and output vector spaces.
During the previous simulations the duty cycles were also calculated,
with a representative set being shown in Figure 6.26, and once again,
comparing this with the results from the Matlab simulations shows that
the control block is working as expected. The stepped nature of the
duty cycle waveforms seen in Figure 6.26 when compared to the Matlab
version shown in Figure 6.14a is because the control block duty cycles
only get calculated once per sampling period, whereas with the Matlab
simulation they were calculated at every iteration step.
With the two different set of internal variables being shown to be correct
for these simulations, along with the duty cycles having been calculated
correctly, the nal stage in the validation of the control block itself is
to show that the output voltage from the converter is being generated as
expected, and equals the demand voltage. Now this was a relatively easy
procedure with the Matlab simulations as the calculated output from
the simulations was a time averaged output signal, which to show it
was working, exactly matched the demand signal. Now that the output
voltage is actually being made up of the space vector modulated output
from the converter, it needs a little more care in checking the results.
Figure 6.27 shows a plot comparing the demand voltage Vdema with the
resulting phase-neutral voltage that is generated by the converter between output legs a and n. This is a valid comparison because the set
of demand voltages only contains the 3 values relating to Va, Vb and
Vc and due to equation 5.3.2 are therefore all implicitly dened with
respect to the neutral leg of the converter. As can be seen, the two
plots are exactly in phase with each other, but due to the switching
138
Vdema =
3
2Vinrms =
2
3
.240 = 293.94V
2
(6.2.5)
Reading this value from Figure 6.29, the magnitude of the fundamental
frequency is shown to be 293.09V. On checking the other two output
phase-neutral voltages, these were also shown to match the phase of
their respective demand voltages, along with having very similar spectra
to that shown in Figure 6.29. From these it is therefore possible to state
that the converter is producing the correct output voltage, and that this
results matches up with the results from the Matlab simulations, thus
proving that the duty cycle calculations and switching state selection are
correct with respect to the output voltage.
Having shown that the simulation is generating the correct output voltages at the terminals to the converter, the resultant voltages which are
then produced across the load resistances were then plotted, and these
can be seen in Figure 6.30. These show each of the voltages to be almost
perfectly sinusoidal, with only a small amount of switching frequency
ripple being evident, and this result is backed up by looking at the plot
of the FFT of one of these voltages which is shown in Figure 6.31.
The clean spectra for these voltages once again re-enforces the evidence
from the spectra and waveforms of the terminal voltages in showing that
this space vector modulation technique is working for the 4-leg matrix
converter. It shows that the state of the converter switches is correct for
each of the required switching states, as the converter is producing the
correct voltages at its terminals in order to give the smooth sinusoidal
outputs across the load.
So looking now at these switching signals from the control block, Figure 6.32 shows how they change over a single clock cycle, and the effect
that they have on the output legs. The gate signals themselves are named
so that the input phase is listed rst, with the output second, so SAb is
139
So, with the output voltage still set at 23 Vinrms , and the load resistance
being set at 30, this gives a peak input current of
I A p eak =
2
1 207.852
Vout
= 2
= 8.48A
Vinrms R
240 30
140
(6.2.6)
(6.2.7)
and it is these which lead to the distortion of the input current waveform.
These low frequency harmonics are extremely undesirable in the input
current, and this shows one of the main limitations of the 4-leg matrix
converter.
Unlike with a balanced load, an unbalanced load on a 3-phase supply
will always draw pulsating power, and because there are no energy storage components within the converter to be able to smooth out the power
141
Vdema =
Vdemb =
Vdemb =
Vin peak
2
Vin peak
4
Vin peak
2
2
3
4
3
(6.2.8)
2240.
6.2.2
Now that a basic set of simulations had been performed, which had
shown that the modulation technique was working correctly, a further
set of simulations were used to look into some of the possible issues
which could arise with the physical implementation of the converter.
6.2.2.1
The rst stage of this investigation was to look into the effects on the
modulation process of the constraints that would placed upon it by the
use of a digital signal processor (DSP) for performing the calculations,
and a eld programmable gate array (FPGA) for controlling the switching of the devices. The actual design of these elements is described in
detail in Chapter 7, but for these simulations a number of assumptions
needed to be made about the implementation. The assumptions made
are as follows:
The switching frequency for the converter would be 12.5kHz
The FPGA clock frequency would be 50MHz
The A/D converters would have be 14 bit resolution
Each of these three assumptions introduces a different limitation into
the previously ideal nature of the simulations, and each of these will be
looked at in turn.
With the current ideal nature of the simulation, at any point in time the
software performs all the necessary calculations it needs at that time
144
147
With the device commutation scheme in place and working, the next
stage in the investigation was to look at the effect that the introduction
of a typical supply inductance has on the operation of the converter. As
mentioned in Section 6.2.2.1 earlier in this chapter, this is especially important during unbalanced operation. This is where the current drawn
from the supply is not purely at the fundamental frequency of the input
149
150
151
P1
c_pwl
P1
Gate1 Sw1_4
P1
Gate1 Sw1_2
Gate2
P1
Gate1 Sw1_1
Gate2
Gate1 Sw1_3
Gate2
P2
3.3u
Gate2
P2
P2
P2
3.3u
ampl
c_pwl
drive_3p
frq
Three Phase
Motor Driver
n
a
b
c
1m
P1
P1
Gate1 Sw1_8
P1
Gate1 Sw1_6
Gate2
P1
Gate1 Sw1_5
Gate2
Gate1 Sw1_7
Gate2
P2
Gate2
P2
P2
P2
3.3u
Voltage
to
Control
Interface
Voltage
to
Control
Interface
clock_l4
LOGIC_4
CLOCK
A
B
C
E
l e
to0
to1
to2
to3
to4
to5
to6
to7
SBx1
SBx2
to0
to1
P1
SAx
SBx
SCx
Gate1 Sw1_10
Gate2
SCx1
SCx2
to2
SAx1
SAx2
CurrDir
SBx1
SBx2
to0
to1
P2
P1
SAx
SBx
SCx
Gate1 Sw1_9
Gate2
SCx1
SCx2
to2
Commutation Control a
T_dead:500n
SAx1
SAx2
CurrDir
Commutation Control b
T_dead:500n
SBx1
SBx2
to0
to1
P2
P1
A
B
C
SAx1
SAx2
CurrDir
Voltage
to
Control
Interface
A
B
C
SAx
SBx
SCx
+
A
B
C
A
B
C
1m
SAx
SBx
SCx
Gate1 Sw1_11
Gate2
SCx1
SCx2
SAx1
SAx2
CurrDir
to2
Commutation Control c
T_dead:500n
SBx1
SBx2
to0
to1
P2
P1
Gate1 Sw1_12
Gate2
SCx1
SCx2
to2
P2
Commutation Control n
T_dead:500n
clk
gnd
Voltage
to
Control
Interface
1m
Voltage
to
Control
Interface
1m
Voltage
to
Control
Interface
1m
Voltage
to
Control
Interface
1m
gnd
5m
2.5m
1m
2.5
5m
10meg
2 LC
(6.2.9)
Figure 6.57 shows what happened to the input voltage when this circuit
was simulated with the following unbalanced load, which was chosen
to ensure that the unbalance not only resulted in giving different load
currents, but also that the displacement factor for each output phase
would also be different.
L a = 5mH
R a = 1
Lb = 2.5mH
(6.2.10)
Rb = 2.5
Lc = 1mH
Rc = 5
As can be seen in Figure 6.57, the input current, and so also the voltage
at the converter terminals is very distorted, and it appears that there is
an interaction between the lter and the converter. In this instance, as
152
154
155
()
156
60.0m
0.0
2.0
4.0
6.0
62.5m
65.0m
67.5m
72.5m
75.0m
77.5m
80.0m
t(s)
82.5m
85.0m
87.5m
90.0m
70.0m
92.5m
95.0m
97.5m
0.1
isector
beta_in_bar
beta_in
(): t(s)
()
157
60.0m
0.0
2.0
4.0
6.0
8.0
10.0
12.0
70.0m
75.0m
80.0m
t(s)
85.0m
90.0m
95.0m
65.0m
0.1
tetra
sign
prism
(): t(s)
()
158
90.0m
0.0
0.2
0.4
0.6
0.8
1.0
91.0m
93.0m
94.0m
95.0m
t(s)
96.0m
97.0m
98.0m
92.0m
99.0m
0.1
delta_t
delta_6
delta_5
delta_4
delta_3
delta_2
delta_1
(): t(s)
(V)
159
Figure 6.27: Plot of Demand and Output Phase-neutral Voltage for Leg a
80.0m 81.0m 82.0m 83.0m 84.0m 85.0m 86.0m 87.0m 88.0m 89.0m 90.0m 91.0m 92.0m 93.0m 94.0m 95.0m 96.0m 97.0m 98.0m 99.0m
t(s)
600.0
300.0
0.0
300.0
600.0
0.1
Vdema
Van
(V): t(s)
(V)
160
90.0m
600.0
400.0
200.0
0.0
200.0
400.0
600.0
91.0m
93.0m
94.0m
95.0m
t(s)
96.0m
97.0m
98.0m
Figure 6.28: Plot of Output Phase Voltage for Leg a referenced to Leg n
92.0m
99.0m
0.1
Van
(V) : t(s)
(V)
161
0.0
0.0
100.0
200.0
300.0
10.0k
15.0k
20.0k
25.0k
f(Hz)
30.0k
35.0k
40.0k
45.0k
50.0k
Figure 6.29: Plot of the FFT of the Output Phase-neutral Voltage on Leg a with respect to Leg n
5.0k
fft(Van)
(V): f(Hz)
(V)
162
82.0m
84.0m
86.0m
88.0m
90.0m
t(s)
92.0m
94.0m
96.0m
98.0m
Figure 6.30: Plot of the 3 Output Phase-neutral Voltages at the load and reference to the load neutral point
80.0m
300.0
200.0
100.0
0.0
100.0
200.0
300.0
0.1
Vcn_load
Vbn_load
Van_load
(V): t(s)
(V)
163
0.0
0.0
100.0
200.0
300.0
10.0k
15.0k
20.0k
25.0k
f(Hz)
30.0k
35.0k
40.0k
45.0k
Figure 6.31: Plot of the FFT of Output Phase-neutral Voltage on Leg a at the load
the
100Hz Demand Frequency- 23 modulation index - 12.5kHz Sampling Frequency
5.0k
50.0k
fft(Van_load)
(V): f(Hz)
164
51.35m
0.0
0.0
0.0
0.0
clk
SCn
SCc
SCb
SCa
SBn
SBc
SBb
SBa
SAn
SAc
SAb
SAa
51.36m
51.38m
51.39m
51.4m
t(s)
51.41m
51.42m
51.43m
51.44m
51.37m
51.45m
Vn
Vc
Vb
Va
(V): t(s)
(V)
165
82.0m
400.0
200.0
0.0
200.0
400.0
SCa
SBa
SAa
82.2m
82.3m
82.4m
82.5m
t(s)
82.6m
82.7m
82.8m
82.9m
Figure 6.33: Plot showing the Switching of the Voltage on Leg a between the Input Phases
82.1m
83.0m
Va
VC
VB
VA
(V) : t(s)
(V)
166
90.0m
400.0
200.0
0.0
200.0
400.0
92.0m
93.0m
94.0m
95.0m
t(s)
96.0m
97.0m
98.0m
99.0m
Figure 6.34: Plot of Output Phase Voltages for Leg a referenced to the supply neutral
91.0m
0.1
Va
(V) : t(s)
(V)
167
400.0
200.0
0.0
200.0
400.0
60.0m
20.0
15.0
10.0
5.0
0.0
5.0
10.0
15.0
20.0
68.0m
72.0m
76.0m
80.0m
t(s)
84.0m
88.0m
92.0m
64.0m
96.0m
100m
VA
(V) : t(s)
IA
(A): t(s)
(A)
(A)
168
0.0
2.0
4.0
6.0
8.0
10.0
0.0
5.0k
15.0k
20.0k
25.0k
f(Hz)
30.0k
35.0k
40.0k
45.0k
Figure 6.36: Plot of the FFT of Input Current for Supply Phase A
10.0k
50.0k
fft(IA)
(A): f(Hz)
(A)
169
0.0
10.0
20.0
30.0
0.0
f(Hz)
10.0k
15.0k
Figure 6.37: Plot of the FFT of Input Current for Supply Phase A scaled as a % of the fundamental
5.0k
fft(IA)
(A): f(Hz)
(A)
170
60.0m
30.0
20.0
10.0
0.0
10.0
20.0
30.0
70.0m
75.0m
80.0m
t(s)
85.0m
90.0m
95.0m
65.0m
0.1
Ic
Ib
Ia
(A) : t(s)
(V)
(V)
(V)
171
0.0
0.0
100.0
200.0
300.0
0.0
100.0
200.0
300.0
0.0
100.0
200.0
300.0
1.0k
3.0k
4.0k
5.0k
6.0k
7.0k
8.0k
f(Hz)
9.0k
10.0k
11.0k
12.0k
13.0k
14.0k
Figure 6.39: Plot of the FFT of Output Phase-neutral Voltages with an Unbalanced Load
2.0k
15.0k
fft(Vcn)
fft(Vbn)
fft(Van)
(V): f(Hz)
(V)
172
400.0
200.0
0.0
200.0
400.0
65.0m
70.0m
75.0m
80.0m
t(s)
85.0m
90.0m
95.0m
Figure 6.40: Plot of the Input Current and Voltage for Supply Phase A with an Unbalanced Load
60.0m
20.0
15.0
10.0
5.0
0.0
5.0
10.0
15.0
20.0
100m
VA
(V) : t(s)
IA
(A): t(s)
(A)
(A)
173
0.0
1.0
2.0
3.0
4.0
5.0
0.0
1.0k
3.0k
4.0k
5.0k
6.0k
7.0k
8.0k
f(Hz)
9.0k
10.0k
11.0k
12.0k
13.0k
14.0k
Figure 6.41: Plot of the FFT of Input Current for Supply Phase A with an Unbalanced Supply
2.0k
(250.0, 0.80407)
(150.0, 0.78378)
15.0k
fft(IA)
(A): f(Hz)
(V)
(V)
(V)
174
0.0
50.0
0.0
100.0
150.0
200.0
0.0
50.0
100.0
150.0
200.0
0.0
50.0
100.0
150.0
200.0
3.0k
4.0k
5.0k
6.0k
7.0k
8.0k
f(Hz)
9.0k
10.0k
11.0k
12.0k
13.0k
14.0k
peak
Vin
15.0k
Figure 6.42: Plot of FFT of the 3 Output Phase-neutral Voltages for Unbalanced Demands
2.0k
1.0k
fft(Vcn)
fft(Vbn)
fft(Van)
(V): f(Hz)
175
0.0
50.0
0.0
100.0
5.0k
10.0k
15.0k
20.0k
f(Hz)
25.0k
30.0k
35.0k
40.0k
45.0k
50.0k
fft(Van)
(V): f(Hz)
Figure 6.43: Plot of the FFT of the Output Phase-neutral Voltage on Leg a with respect to Leg n using the Two Zero Switching
Sequence
100Hz Demand Frequency- q = 0.5 - 12.5kHz Sampling Frequency
(V)
150.0
176
0.0
50.0
0.0
100.0
5.0k
10.0k
15.0k
20.0k
25.0k
f(Hz)
30.0k
35.0k
40.0k
45.0k
50.0k
fft(Van)
(V): f(Hz)
Figure 6.44: Plot of the FFT of the Output Phase-neutral Voltage on Leg a with respect to Leg n using the Three Zero Switching
Sequence
100Hz Demand Frequency- q = 0.5 - 12.5kHz Sampling Frequency
(V)
150.0
177
0.5
400.0
200.0
0.0
200.0
0.502
0.504
0.506
0.508
0.51
t(s)
0.512
0.514
0.516
0.518
0.52
VA Sampled
VA Input
(V) : t(s)
Figure 6.45: Plot of the Supply Voltage alongside the Sampled Voltage showing the Processing Delay within the Converter
50Hz Input Frequency - 12.5kHz Sampling Frequency
(V)
400.0
(V)
178
0.507
0.508
0.509
0.51
t(s)
0.511
0.512
0.513
0.514
VA Calc Delta
VA Input
Figure 6.46: Plot of the Supply Voltage alongside the Sampled Voltage showing the Processing Delay Compensation
50Hz Input Frequency - 12.5kHz Sampling Frequency
0.506
400.0
200.0
0.0
200.0
400.0
(V) : t(s)
179
0.0
0.0
50.0m
0.1
0.15
0.2
0.25
0.0
50.0m
0.1
0.15
0.2
100.0 200.0 300.0 400.0 500.0 600.0 700.0 800.0 900.0 1.0k 1.1k
f(Hz)
1.2k
1.3k
1.4k
1.5k
1.6k
1.7k
1.8k
1.9k
2.0k
% : f(Hz)
fft(Van) No comp
% : f(Hz)
Figure 6.47: Plot of the FFT of the Output Voltage, Van, with and and without compensation for the processing delay using the
Delta method
100Hz Output Frequency - 12.5kHz Sampling Frequency
0.25
(V)
(V)
180
0.502
0.504
0.506
0.508
0.51
t(s)
0.512
0.514
0.516
0.518
0.52
VA PLL
(V) : t(s)
VA Noisy
(V) : t(s)
Figure 6.48: Plot of the noisy Input Voltage, VA, alongside the PLL tracked fundamental frequency output used by the
converter
50Hz Input Frequency - 12.5kHz Sampling Frequency
0.5
500.0
0.0
500.0
500.0
0.0
500.0
181
0.0
0.5
150.0
100.0
50.0
0.0
50.0
100.0
150.0
0.0
0.2
0.4
0.6
0.8
1.0k
0.505
2.0k
0.51
3.0k
4.0k
0.515
5.0k
0.52
6.0k
7.0k
0.525
t(s)
8.0k
f(Hz)
9.0k
0.53
10.0k
0.535
11.0k
12.0k
0.54
13.0k
15.0k
0.545
14.0k
0.55
(V) : t(s)
16.0k
% : f(Hz)
Figure 6.49: Plot of the Output Voltage, Van, and its Spectra showing the effect of Noise when using the Delta method of delay
compensation
100Hz Output Frequency - 12.5kHz Sampling Frequency
(V)
1.0
(V)
0.0
182
1.0k
0.505
2.0k
0.51
3.0k
4.0k
0.515
5.0k
0.52
6.0k
7.0k
0.525
t(s)
8.0k
f(Hz)
9.0k
0.53
10.0k
0.535
11.0k
12.0k
0.54
13.0k
15.0k
0.545
14.0k
0.55
(V) : t(s)
16.0k
% : f(Hz)
Figure 6.50: Plot of the Output Voltage, Van, and its Spectra showing the effect of Noise when using the PLL for delay
compensation
100Hz Output Frequency - 12.5kHz Sampling Frequency
0.5
150.0
100.0
50.0
0.0
50.0
100.0
150.0
0.0
0.2
0.4
0.6
0.8
1.0
183
0.0
0.0
50.0m
0.1
0.15
0.2
0.25
0.0
50.0m
0.1
0.15
0.2
100.0 200.0 300.0 400.0 500.0 600.0 700.0 800.0 900.0 1.0k 1.1k
f(Hz)
1.2k
1.3k
1.4k
1.5k
1.6k
1.7k
1.8k
1.9k
2.0k
fft(Van) FPGA
% : f(Hz)
fft(Van) No comp
% : f(Hz)
Figure 6.51: Plot of the FFT of the Output Voltage, Van, with and without the truncation effect of the FPGA minimum step size
100Hz Output Frequency - 12.5kHz Sampling Frequency
0.25
184
0.0
0.0
50.0m
0.1
0.15
0.2
0.25
0.0
50.0m
0.1
0.15
0.2
100.0 200.0 300.0 400.0 500.0 600.0 700.0 800.0 900.0 1.0k 1.1k
f(Hz)
1.2k
1.3k
1.4k
1.5k
1.6k
1.7k
1.8k
1.9k
2.0k
% : f(Hz)
fft(Van) No comp
% : f(Hz)
Figure 6.52: Plot of the FFT of the Output Voltage, Van, showing the original along with the output which is compensated for
the truncation effect of the FPGA minimum step size
100Hz Output Frequency - 12.5kHz Sampling Frequency
0.25
(V)
185
0.0
0.51
0.515
0.52
0.505
1.0k
f(Hz)
0.525
t(s)
1.1k
1.2k
0.53
1.3k
1.4k
0.535
1.5k
1.6k
0.54
1.7k
1.8k
0.545
1.9k
2.0k
fft(Van)
% : f(Hz)
0.55
Van
(V) : t(s)
Figure 6.53: Plot of the Output Voltage, Van, and its Spectra, showing the combined effect of the different types of
compensation
100Hz Output Frequency - 12.5kHz Sampling Frequency
0.0
50.0m
0.1
0.15
0.2
0.25
0.5
150.0
100.0
50.0
0.0
50.0
100.0
150.0
SCa2
SCa1
SBa2
SBa1
SAa2
SAa1
SBa
SAa
186
0.50015
400.0
200.0
0.0
200.0
400.0
0.5001505
0.500151
0.5001515
0.500152
t(s)
0.5001525
0.500153
0.5001535
0.500154
VB Input
VA Input
Va Output
(V) : t(s)
Figure 6.54: Plot showing the switching sequence used for commutation going from a lower to a higher voltage input phase
(V)
SCa2
SCa1
SBa2
SBa1
SAa2
SAa1
SCa
SBa
187
0.500142
400.0
200.0
0.0
200.0
400.0
0.5001425
0.500143
0.5001435
0.500144
t(s)
0.5001445
0.500145
0.5001455
0.500146
VC Input
VB Input
Va Output
(V) : t(s)
Figure 6.55: Plot showing the switching sequence used for commutation going from a lower to a higher voltage input phase
(V)
(V)
188
0.0
0.0
50.0m
0.1
0.15
0.2
0.25
0.5
150.0
100.0
50.0
0.0
50.0
100.0
150.0
0.51
0.515
0.52
1.0k
f(Hz)
0.525
t(s)
1.1k
1.2k
0.53
1.3k
1.4k
0.535
1.5k
1.6k
0.54
1.7k
1.8k
0.545
1.9k
Figure 6.56: Plot of the FFT of the Output Voltage, Van, comparing the commutated output to the ideal
100Hz Output Frequency - 12.5kHz Sampling Frequency
0.505
2.0k
fft(Van)
% : f(Hz)
0.55
Van - Com
(V) : t(s)
189
(A)
0.5
1.0k
0.0
1.0k
500.0
0.0
500.0
0.505
0.51
0.515
0.52
0.525
t(s)
0.53
0.535
0.54
0.545
0.55
VA Converter
(V) : t(s)
IA Supply
(A) : t(s)
VA Supply
(V) : t(s)
Figure 6.57: Plot of Input Phase A Supply Voltage and Current, along with the voltage seen at the converter, when using a
1mH/33F Input Filter with no PLL
50Hz Input Frequency - 12.5kHz Sampling Frequency
100.0
0.0
(V)
(V)
100.0
190
40.0
20.0
0.0
0.5
500.0
0.0
500.0
500.0
0.0
500.0
0.505
0.51
0.515
0.52
0.525
t(s)
0.53
0.535
0.54
0.545
0.55
VA Converter
(V) : t(s)
IA Supply
(A) : t(s)
VA Supply
(V) : t(s)
Figure 6.58: Plot of Input Phase A Supply Voltage and Current, along with the voltage seen at the converter, when using a
1mH/47F Input Filter with no PLL
50Hz Input Frequency - 12.5kHz Sampling Frequency
(A)
20.0
(V)
(V)
40.0
191
(V)
0.5
500.0
0.0
500.0
20.0
0.0
20.0
0.505
0.51
0.515
0.52
0.525
t(s)
0.53
0.535
0.54
0.545
0.55
VA Converter
(V) : t(s)
IA Supply
(A) : t(s)
VA Supply
(V) : t(s)
Figure 6.59: Plot of Input Phase A Supply Voltage and Current, along with the voltage seen at the converter, when using a
1mH/33F Input Filter using a PLL
50Hz Input Frequency - 12.5kHz Sampling Frequency
500.0
0.0
(A)
(V)
500.0
(V)
Mag(V/Hz)
192
0.0
1.0k
2.0k
0.505
3.0k
0.51
4.0k
5.0k
0.515
6.0k
0.52
7.0k
8.0k
f(Hz)
0.525
t(s)
9.0k
0.53
10.0k
11.0k
0.535
12.0k
0.54
13.0k
14.0k
0.545
15.0k
fft(Van)
Mag(V/Hz) : f(Hz)
0.55
Van
(V) : t(s)
Figure 6.60: Plot of Output Voltage, Van, along with its Spectra, when using a 1mH/33F Input Filter using a PLL
100Hz Output Frequency - 12.5kHz Sampling Frequency
0.0
10.0
20.0
30.0
40.0
50.0
0.5
60.0
40.0
20.0
0.0
20.0
40.0
60.0
193
(A)
0.15
400.0
200.0
0.0
200.0
400.0
400.0
200.0
0.0
200.0
400.0
0.175
0.2
0.225
0.25
0.275
0.3
0.325
0.35
t(s)
0.375
0.4
0.425
0.45
0.475
0.5
0.525
0.55
VA Converter
(V) : t(s)
IA Supply
(A) : t(s)
VA Supply
(V) : t(s)
Figure 6.61: Plot of Input Phase A Supply Voltage and Current, along with the voltage seen at the converter, with the peak
input voltage decreasing from 340V to 250V
50Hz Input Frequency - 12.5kHz Sampling Frequency
20.0
0.0
(V)
(V)
20.0
(V)
194
0.0
25.0
50.0
75.0
0.0
100.0
125.0
150.0
0.15
150.0
100.0
50.0
0.0
50.0
100.0
0.2
0.225
0.25
0.275
0.3
0.325
0.175
1.0k
f(Hz)
0.35
t(s)
1.1k
0.375
1.2k
0.4
1.3k
1.4k
0.425
1.5k
0.45
1.6k
0.475
1.7k
0.5
1.8k
1.9k
0.525
2.0k
fft(Van)
(V) : f(Hz)
0.55
Van
(V) : t(s)
Figure 6.62: Plot of Output Voltage, Van, along with its Spectra, with the peak input voltage decreasing from 340V to 250V
100Hz Output Frequency - 12.5kHz Sampling Frequency
(V)
150.0
195
(A)
0.15
400.0
200.0
0.0
200.0
400.0
400.0
200.0
0.0
200.0
400.0
0.175
0.2
0.225
0.25
0.275
0.3
0.325
0.35
t(s)
0.375
0.4
0.425
0.45
0.475
0.5
0.525
0.55
VA Converter
(V) : t(s)
IA Supply
(A) : t(s)
VA Supply
(V) : t(s)
Figure 6.63: Plot of Input Phase A Supply Voltage and Current, along with the voltage seen at the converter, with the input
frequency increasing from 50Hz to 100Hz
12.5kHz Sampling Frequency
20.0
0.0
(V)
(V)
20.0
(V)
196
0.0
25.0
50.0
75.0
0.0
100.0
125.0
150.0
0.15
150.0
100.0
50.0
0.0
50.0
100.0
0.2
0.225
0.25
0.275
0.3
0.325
0.175
1.0k
f(Hz)
0.35
t(s)
1.1k
0.375
1.2k
0.4
1.3k
1.4k
0.425
1.5k
0.45
1.6k
0.475
1.7k
0.5
1.8k
1.9k
0.525
2.0k
fft(Van)
(V) : f(Hz)
0.55
Van
(V) : t(s)
Figure 6.64: Plot of Output Voltage, Van, along with its Spectra, with the input frequency increasing from 50Hz to 100Hz
100Hz Output Frequency - 12.5kHz Sampling Frequency
(V)
150.0
197
(A)
0.15
400.0
200.0
0.0
200.0
400.0
400.0
200.0
0.0
200.0
400.0
0.175
0.2
0.225
0.25
0.275
0.3
0.325
0.35
t(s)
0.375
0.4
0.425
0.45
0.475
0.5
0.525
0.55
VA Converter
(V) : t(s)
IA Supply
(A) : t(s)
VA Supply
(V) : t(s)
Figure 6.65: Plot of Input Phase A Supply Voltage and Current, along with the voltage seen at the converter, with the peak
input voltage decreasing from 340V to 250V while the input frequency increases from 50Hz to 100Hz
12.5kHz Sampling Frequency
20.0
0.0
(V)
(V)
20.0
(V)
198
0.0
25.0
50.0
75.0
0.0
100.0
125.0
150.0
0.15
150.0
100.0
50.0
0.0
50.0
100.0
0.2
0.225
0.25
0.275
0.3
0.325
0.175
1.0k
f(Hz)
0.35
t(s)
1.1k
0.375
1.2k
0.4
1.3k
1.4k
0.425
1.5k
0.45
1.6k
0.475
1.7k
0.5
1.8k
1.9k
0.525
2.0k
fft(Van)
(V) : f(Hz)
0.55
Van
(V) : t(s)
Figure 6.66: Plot of Output Voltage, Van, along with its Spectra, with the peak input voltage decreasing from 340V to 250V
while the input frequency increases from 50Hz to 100Hz
100Hz Output Frequency - 12.5kHz Sampling Frequency
(V)
150.0
199
40.0
20.0
0.0
0.15
500.0
0.0
500.0
500.0
0.0
500.0
0.175
0.2
0.225
0.25
0.275
0.3
0.325
0.35
t(s)
0.375
0.4
0.425
0.45
0.475
0.5
0.525
0.55
VA Converter
(V) : t(s)
IA Supply
(A) : t(s)
VA Supply
(V) : t(s)
Figure 6.67: Plot of Input Phase A Supply Voltage and Current, along with the voltage seen at the converter, with the peak
input voltage decreasing from 340V to 250V while the input frequency increases from 50Hz to 100Hz, driving an unbalanced
load
12.5kHz Sampling Frequency
(A)
20.0
(V)
(V)
40.0
(V)
200
0.0
10.0
20.0
30.0
40.0
50.0
60.0
0.0
0.15
75.0
50.0
25.0
0.0
25.0
50.0
0.2
0.225
0.25
0.275
0.3
0.325
0.175
1.0k
f(Hz)
0.35
t(s)
1.1k
0.375
1.2k
0.4
1.3k
1.4k
0.425
1.5k
0.45
1.6k
0.475
1.7k
0.5
1.8k
1.9k
0.525
2.0k
fft(Van)
(V) : f(Hz)
0.55
Van
(V) : t(s)
Figure 6.68: Plot of Output Voltage, Van, along with its Spectra, with the peak input voltage decreasing from 340V to 250V
while the input frequency increases from 50Hz to 100Hz, driving an unbalanced load
75Hz Output Frequency - 12.5kHz Sampling Frequency
(V)
75.0
C HAPTER 7
3/2Vin
201
Power Circuit
FPGA
DSP
Control
PC
Load
7.2
Gate signals
from FPGA
Gate Drive
Circuit
Gate Drive
Circuit
Gate Drive
Circuit
Gate Drive
Circuit
Gate Drive
Circuit
Gate Drive
Circuit
Gate Drive
Circuit
Gate Drive
Circuit
Gate Drive
Circuit
Voltage
Sense
Gate Drive
Circuit
Gate Drive
Circuit
Gate Drive
Circuit
415V
Voltage
Sense
Sense Signals
to FPGA
Current Sense
and direction
Current Sense
and direction
Current Sense
and direction
Current Sense
and direction
Load
7.2.1
Devices
204
7.2.2
205
Ux+15V
+5V
U1GND
GND
UxGND
Gate_signal_a
Gate drive
to switch a
GND
GND
U1GND
U1GND
Ux-15V
Ux-15V
UxGND
Ux+15V
+5V
Ux+15V
+5V
U1GND
GND
UxGND
Gate_signal_b
Gate drive
to switch b
GND
GND
U1GND
U1GND
Ux-15V
Ux-15V
UxGND
Figure 7.4: Circuit Diagram of the Gate Drive Circuit for a single
Bi-directional Switch
For the gate drive circuit for this project, the IGBT gate is driven from
+15V and -15V sources. Figure 7.4 shows the circuit diagram for a gate
drive for a single bi-directional switch. As can be seen there are a pair
of bipolar transistors which are driven in anti-phase to switch the gate
circuit between the two voltages, and the gate resistor is used to control
the maximum current fed into/drawn from the gate.
Alongside the specic requirements for driving the IGBT itself, the matrix converter also has specic requirement for its gate drives, and that is
that each of the gate drives needs to be isolated. This is because the gate
drive circuits must always reference the emitter voltage of the devices,
206
7.2.3
As this converter is going to be utilising the 4-step commutation sequence, as examined in Section 4.3 above, the FPGA, which generates the
control signals, needs to know at any switching instant which direction
the current is owing through each bi-directional switch. To achieve
this a current direction sensing circuit is built into the output of each
switching leg, and is made up of a pair of anti-parallel diodes. Each
pair of anti-parallel diodes are used to monitor the current direction in
their particular switching leg, with the voltage across the diodes being
positive for current ow in one direction, and negative in the other depending on which diode is conducting. This voltage is then fed into a
comparator to produce a logic level signal that is capable of being fed
back, through an isolation barrier, to the FPGA. This logic level is then
used to determine the commutation sequence required when switching
between output states. By using the diodes in this way it is possible to
detect down to very low current values so ensuring that the optimum
commutation sequence is used. Below these current values where the
circuit stops working, the current levels are so small that it makes no
real difference which commutation sequence is used.
Alongside the current direction sensing, there is also an inline current
transducer which is used for monitoring the output current from each
converter leg. This uses a device from a company called LEM which
has an isolated output which can then be fed directly out to the A/D
converter circuits on the FPGA card. These signals are then used by the
DSP to primarily protect the power circuit against any prolonged overcurrent conditions, although in more complex control schemes these currents can be used as part of the control loop.
207
Isolated
+5V
xcd+15V
+5V
Vcc
Current
direction
Vo
Vgnd
xcd-15V
GND
GND
Current
transducer
CTu
+15
+15V
-15
-15V
Output
to
load
7.2.4
In order for Space Vector Modulation to work it is vital to know both the
direction and phase of the input voltage space vector, and to do this we
need to know each of the three input voltages. This is achieved in this
instance by the use of LEM voltage transducers. Even though the values
of all three input phases needs to be known, from the circuit diagram in
Figure 7.6, it can be seen that there are only two voltage transducers, and
these are being used to measure the two line-to-line voltages VAB and
VBC . The reason that the voltage sensing is performed in this manner is
that there is no neutral line at the input side of the converter, as the idea
behind this type of converter is to be able to generate the 3-phase plus
neutral from a balanced 3-phase supply.
So, as it is known that the input phases will form a three phase set where
VA + VB + VC = 0
(7.2.1)
(7.2.2)
Voltage
transducer
Vrs
+15
+15V
-15
-15V
S'meas
Smeas
Smeas
Voltage
transducer
Vst
+15
+15V
-15
-15V
T'meas
Tmeas
VBC = VB VC
(7.2.3)
(7.2.4)
(7.2.5)
1
(2VAB VBC )
3
(7.2.6)
From 7.2.6 it can be seen that by just knowing these two line-to-line
voltages it is possible to calculate one of the line-to-neutral voltages, and
from there the other two line-to-neutral voltages are trivial to solve, such
that
VA
VB =
VC
1
3
(2VAB VBC )
VA VAB
VA VB
209
(7.2.7)
7.2.5
Clamp Circuit
This is a relatively simple protection circuit which gives the input and
output currents a freewheel path in the case of a device not being switched
correctly, and is the parallel diode bridge as discussed in Section 6.2.2.3.
While it helps the converter during normal commutation, it also provides some protection for the matrix converter if problems arise a commutation by operating as a voltage clamp. The circuit layout for this
is shown in Figure 7.7, and as can be seen, it is a pair of back to back
diode rectiers, one connected to the input and the other to the output.
The rectied output from these bridges is then connected to a pair of
series capacitors. If one of the switches fails for some reason, then any
current in that input or output phase will be able to ow through its
respective diode in the clamp and then into/through the capacitor. This
stops any large increase in voltage which could occur due to inductances
within the input and output circuits if an output leg was allowed to go
open circuit. If this is just a problem with a single commutation the
time would be small and so the associated voltage increase would also
be small, but if there is a larger problem then the increase in capacitor
voltage can then be used to trigger a trip within the control circuit and
the converter can be stopped. Any currents still owing in the input and
output legs can then be allowed to safely freewheel through the clamp
circuit. The associated resistor is there to provide a discharge path for the
capacitor, so that when the converter is tripped the freewheeling energy
can be safely dissipated.
Included in this circuit is another LEM voltage transducer which is used
to sense the clamp circuit voltage, the output of which is fed into a comparator. Once the clamp voltage rises above the comparators reference
voltage the output goes high and this signal is fed back into the FPGA
board and thus the DSP to enable it to stop the converter.
7.2.6
This is a multi-layer PCB which along with the top and bottom signal
layers also has four internal layers, three of which are shared between
the three input and three output phases, with an input and output phase
210
M
+15
3-Phase
Input from
mains supply
+15V
-15
-15V
4-Phase
Input from
Converter output
DC Clamp-
+5V
+5V
GND
GND
GND
GND
Figure 7.7: Circuit Diagram of the Voltage Clamp used for Converter
Protection
per layer, and then another layer to allow a separate ground plane to
be used. By using the internal layers in this way, it allows as large an
amount of copper as possible to be used for connecting the phases to
and from the devices. There is the obvious advantage with having as
much copper as possible as it keeps the track resistance down and allows
higher power levels to be used, but this also has the additional benet
of allowing the tracks to be as wide as possible and this helps to reduce
the stray inductance in the design. While having any stray inductance
in the output circuit is not really an issue due to the inductors on each of
the output phases, keeping the inductance of the input circuits as low as
possible is an advantage. This is because of the nature of the switching,
where only two input phases are usually conducting, and so the input
currents are discontinuous. As such any stray inductance could cause
spikes on the input voltage.
The layout for this board is an extension of the layout for the 3x3 matrix
circuit. This layout was arranged with the input phases being connected
to the switches in columns down the board, and had the output legs
connected across the board. This arrangement was kept, and allowed
the 4th leg to be placed along the bottom edge of the PCB.
7.2.7
Input Filter
Supply
Converter
L
C
Figure 7.8: Circuit Diagram of the Supply and Input Filter for the 4-Leg
Matrix Converter
Choosing a cut-off frequency at this point means that it is still well below
the switching frequency being used and so be able to effectively lter
it, while also trying to keep the phase shift of the lter as close to zero
over the entire input frequency range, up to 400Hz. This last point is
important as it allows the converter to have a input displacement factor
as close to unity as possible.
7.2.8
The nal part of the power section is the load, and this itself is broken
up into two parts, the output inductors and the load resistance.
For a matrix converter to operate it ideally requires some inductance on
each of the 4 output legs, thus giving it the current stiff characteristic.
The actual inductors used for the converter are a set of 8mH 3-phase
inductors attached to the three main output legs, with a pair of 4mH
inductors connected in series on the neutral leg. This arrangement was
limited in this way because there were no 4-phase inductors available
within the PEMC test lab, as 4-phase output inductors are relatively
212
7.3
7.3.1
On external interrupt
from the FPGA(12.5kHz)
matrix.c
interface.c
Initialise
communications
Read PC Message
Process PC Message
Start Drive
Stop Drive
Change Demand
Capture Data
Look up switching
states
Read/Write
FPGA
Figure 7.9: Block Diagram for the DSP Software for the 4-Leg Matrix
Converter
The software for the DSP is broken up in the 3 main parts, as can be seen
in Figure 7.9, with these being the:
Initialisation and peripheral setup
Interface and main control loop
Matrix SVM interrupt code
The initialisation and peripheral setup code block is the rst block of
code to be run by the DSP when the application starts. This is a straightforward set of functions which are used to initialise and set up the various parts of the DSP and its evaluation board, initialising the main
variables and then setting up the FPGA. This code is just run at the start
of the program, and its nal task is to call the AwaitPCMessage which is
the main communication and control loop.
The communication and control loop is the part of the DSP software
which, as its name suggests, is where theDSP handles its communication
214
7.3.2
A/D Converter
Control and Interface
Hardware trips
Trip
Monitor
Single Leg
4-Step Commutation
Block
PWM
Module
FIFO
FPGA/DSP Interface
Controller
Data Bus
Watchdog
Timer
Single Leg
4-Step Commutation
Block
DSP
Single Leg
4-Step Commutation
Block
Address
Bus
Single Leg
4-Step Commutation
Block
I 2C
Interface
Hardware trip
sensors
Current direction
input from converter
Figure 7.10: Block Diagram showing the main components of the FPGA
An A/D Converter module which controls the A/D converters,
and then demultiplexes the resulting data
A Watchdog timer that will trip the converter if required
A set of Hardware and software trips
The overall design of the FPGA is such that it is seen by the DSP as
a series of memory addresses, which it is able to read from and write
to, and this interface is briey described in Appendix C. By having the
interface arranged in this way, it allows the registers and functions of the
FPGA to be accessed directly by the DSP, this is greatly advantageous as
it means that no special routines are required for reading from or writing
to it, and especially so with respect to the A/D data.
The core functions of the FPGA are set up by the DSP on initialisation,
the most important here being the PWM sampling period, as this controls the interrupt that is sent to the DSP, along with providing all the
timing information for the 4 different commutation blocks. Once the
sampling period is set, the various parts of the FPGA can be initialised,
the outputs turned on, and the FPGA at this point is ready to go.
The most important function for the FPGA is that of generating the gate
control signals which go straight to the gate drive circuits on the power
board. The sequence for this originates in the previous sampling period
where the DSP has calculated the switching states and times for this
sampling period, and this data has been loaded into the FIFO buffer in
the FPGA. Then, as the new sampling period starts, the rst switching
state out of the FIFO buffer gets output by the PWM module, which is
218
219
7.3.3
7.3.4
Matlab/PC
To enable some control over the DSP, and thus the converter, a Matlab
graphical user interface (GUI) was designed that would allow the PC to
stop and start the converter, reset it, and also change the demand voltage
and frequency. During the development of the DSP software, it became
apparent that due to the complexity of the calculations, it would be
difcult to verify if the DSP was actually working correctly, so alongside
221
Figure 7.11: The Main Control Interface for the 4-Leg Matrix Converter
The second window in the Matlab interface, which is shown in Figure 7.12. is used to show information about the state of the FPGA registers, the outputs from the A/D converters, and of a number of other
internal DSP variables. This data is regularly sampled by the PC, and so
this window automatically updates its values.
222
223
C HAPTER 8
Experimental Results
This chapter describes the testing which was performed on the 4-Leg
Matrix Converter described in the previous chapter. The initial testing
was all designed around ensuring that the converter was operational,
before moving on to use the converter to drive a real load, both balanced
and unbalanced.
8.1
Initial Testing
224
8.2
Balanced Load
Once initial testing had been completed, the converter was connected
to a 415V supply, through a variac, and then attached to its load, made
up of 4 8mH inductors and a star connected set of 3 10 resistors as
described in the previous chapter. The following results show results
show that the converter is working as expected.
Due to the trouble with the diode protection circuit, the converter was
being run with an input voltage of approximately 80Vrms , as such the demand voltage was set to a gure of 50Vrms and with a demand frequency
of 100Hz. The results for this test follow.
Figure 8.1 shows the set of output voltages measured across the load
resistances.
Figure 8.3: Plot of the FFT of the Output Phase-neutral Voltage on Leg a
at the Load
50Vrms Demand Voltage - 100Hz Demand Frequency - 12.5kHz
Sampling Frequency
Figure 8.4 shows the an area of the FFT in Figure 8.3 which has been
zoomed in vertically 25 times. As can be seen, the switching frequency
226
Figure 8.5: Plot of the FFT of the Output Phase-neutral Voltage on Leg a
at the Converter
50Vrms Demand Voltage - 100Hz Demand Frequency - 12.5kHz
Sampling Frequency
229
Figure 8.9: Plot of the 4-Leg Matrix Converter Input Current and
Voltage for Phase A
50Vrms Demand Voltage - 100Hz Demand Frequency - 12.5kHz
Sampling Frequency
internal state of the converter. This was originally designed to be able
to check on the correct operation, but it also allows the later storage and
plotting of the results. For these results one of the most interesting of
these is the sampled output current data for each of the three phase legs
a, b and c, with the neutral current being calculated as the sum of the
other 3 legs. Figure 8.10 shows a plot of these currents, and it can be
seen here that the current ripple which is present in the input phases,
looks to also be present here as well. This ties up with a possible problem
with how the converter is behaving around the interfaces between input
sectors and tetrahedrons, and is a problem with the implementation in
this converter rather than a problem with the theory itself.
Also of interest is the plot of the duty cycles shown in Figure 8.11. These
are a snapshot of the duty cycles from the above testing, and as can be
seen, they overall, follow the general shape of the duty cycles seen in
both the Matlab(Figure 6.15a) and Saber(Figure 6.26) simulations. The
slight differences in shape are entirely due to the relative phasing between the input frequency and the output frequency.
And nally, before moving on to look at the behaviour of the converter
under an unbalanced load condition, is the plot shown in Figure 8.12.
This shows the time it takes for the DSP to perform all the calculations
required to determine the switching vectors and times to control the matrix converter, Ti , and plots this value alongside the total time available
230
Figure 8.10: Plot of the 4-Leg Matrix Converter Output Leg Currents for
Balanced Operation
50Vrms Demand Voltage - 100Hz Demand Frequency - 12.5kHz
Sampling Frequency
Figure 8.11: Plot of the 4-Leg Matrix Converter Input Current and
Voltage for Phase A
50Vrms Demand Voltage - 100Hz Demand Frequency - 12.5kHz
Sampling Frequency
within the SVM switching period, Ts . As can be seen, the interrupt
routine takes approximately 60s to perform its calculations, leaving
around 20s left over in which to perform all its other duties. In this case
that time was taken up with communicating with the host PC and performing any commands, however in a real application this time would
also need to be used to implement whatever control strategy was being
231
Figure 8.12: Plot showing Ti , the time taken to process the DSP
interrupt routine, alongside the SVM Period, Ts
232
8.3
Unbalanced Load
Figure 8.14: Plot of the FFT of the Input Current of Phase A with an
Unbalanced Load
50Vrms Demand Voltage - 100Hz Demand Frequency - 12.5kHz
Sampling Frequency
Figure 8.15: Close-up of the FFT of the Input Current of Phase A with
an Unbalanced Load
50Vrms Demand Voltage - 100Hz Demand Frequency - 12.5kHz
Sampling Frequency
Looking now at the voltage that is produced across the load, and as
with the Saber simulations, the voltage waveforms were seen to be sinusoidal, which is shown in Figure 8.17 which shows the FFT of the
output voltage for leg c across the load resistor. Looking more closely at
the lower frequency range in Figure 8.18 once again shows there is very
234
Figure 8.16: Plot of the FFT of the Output Phase-neutral Voltage for Leg
c at the Converter for an Unbalanced Load
50Vrms Demand Voltage - 100Hz Demand Frequency - 12.5kHz
Sampling Frequency
little low frequency distortion, and once again matches very closely with
the results found in the Saber simulations, and shown in Fig 6.39.
Figure 8.17: Plot of the FFT of the Output Phase-neutral Voltage for Leg
c across the Load for an Unbalanced Load
50Vrms Demand Voltage - 100Hz Demand Frequency - 12.5kHz
Sampling Frequency
As with the testing of the balanced load above, the DSP was used to
capture snapshots of the converters operation, and by looking at the
235
Figure 8.19: Plot of the 4-Leg Matrix Converter Output Leg Currents for
Unbalanced Load
50Vrms Demand Voltage - 100Hz Demand Frequency - 12.5kHz
Sampling Frequency
Unfortunately, at this point the converter suffered another failure within
236
8.4
Summary
While the converter was working the experimental results which were
achieved were very good, even though the converter was having to operate at reduced voltages levels. These results matching up with the
simulated results to a high degree, however with the problems in the
diode bridge this did limit the testing, and even cut it short of the ideal.
However, with the results that have been achieved, taking these in conjunction with the simulation results it is reasonable to state that the converter is operating as expected. As such this proves the theory behind
the operation, and shows that the equations derived for calculating the
duty cycles and switching states are correct. As such meets the main
objective set for this work.
Unfortunately, what these results do not prove is whether the actual implementation of this converter is correct as it was just prior to these tests
that it failed once again. Having stated that, during the initial testing
the DSP was put through a large number of different tests, involving a
range of different waveforms, and although not documented here, as no
results were taken during this development work, this did show that the
DSP software was working.
237
C HAPTER 9
Conclusions
The work presented in this thesis has set out to demonstrate the proof of
the theory behind the implementation of Space Vector Modulation on a
4-Leg Matrix converter, and it is believed that with the work presented
here it has been able to do this. The aims and objectives at the start of the
project were simple, to investigate the use of space vector modulation
with the new 4-leg matrix converter, and then to derive its theory and
operation. This theory was then to be tested by both simulation and by
the construction of a demonstration converter.
This investigation started with a look at how space vector modulation
works, and has been implemented, on two related circuits, the 4-leg
inverter and matrix converter, from which the 4-leg matrix converter derives. The work then went on to show a derivation of the theory behind
the 4-leg matrix converter in Chapter 5. This derivation started with
by looking at the input current and output voltages spaces separately,
before being able to combine them in a fashion that allowed both to be
independently controlled with respect to one another.
Having completed this proof, the simulation work then became the next
step, with this forming a large quantity of the work involved in this
paper. The initial simulation work went on to rst prove itself to be
an adequate method, before moving on to demonstrate that the theory
worked perfectly, and the results then provided by further circuit simulation backed these results up further. The simulation work then moved
onto looking at more advanced concepts using the Saber package, and
with these simulations it was possible to show how the converter dealt
with a number of issues around the physical implementation of the converter.
The focus of the work then turned on to building a converter to practically demonstrate these results, by expanding on an existing design
238
C HAPTER 9: C ONCLUSIONS
of converter. While this work resulted in a converter which was capable of operation, a problem which caused a number of failures, limited the experimental results available. However it is felt that despite
these problems with the experimental converter, the results that were
obtained were more than capable of backing up the simulation results
and showing that the theory of operation for the converter was correct.
Since this work a 4-Leg Matrix converter using the derived technique
has been built and proven to be operational[36]. This evidence can only
be strengthened by papers[30, 36] which have been published since this
initial theory was published[32].
At the start of this project there had been no research published into the
operation or control of the 4-Leg Matrix converter, however interest is
starting to be seen with the number of published works increasing as
different aspects of the 4-Leg Matrix converter are investigated[29, 34
41]. It is therefore hoped that the work presented here can be seen as
being part of the beginnings of this interest.
So, despite the problems which were encountered with the build of the
matrix converter, there is sufcient evidence presented here that it is
possible to state that the derivation and implementation of Space Vector
Modulation in a Four-Leg Matrix Converter shown here is correct, and
so the work has been a success.
239
A PPENDIX A
240
A PPENDIX B
IGBT Datasheet
241
SK 60GM123
Absolute Maximum Ratings
Symbol Conditions
IGBT
Values
Units
SEMITOP 2
IGBT Module
SK 60GM123
Inverse Diode
Module
Preliminary Data
Features
Characteristics
Symbol Conditions
IGBT
min.
typ.
max.
Units
Typical Applications*
GM
27-06-2007 DIL
by SEMIKRON
SK 60GM123
Characteristics
Symbol Conditions
Inverse Diode
min.
typ.
max.
Units
SEMITOP 2
IGBT Module
SK 60GM123
Preliminary Data
Features
Typical Applications*
GM
27-06-2007 DIL
by SEMIKRON
SK 60GM123
27-06-2007 DIL
by SEMIKRON
SK 60GM123
27-06-2007 DIL
by SEMIKRON
SK 60GM123
UL recognized file
no. E 63 532
27-06-2007 DIL
by SEMIKRON
A PPENDIX C
247
DPR0 0xA0000000;
31
30
29
15
14
13
PRS
PEN
PEN
CIP
Read
Write
Read
Write
PRS
28
27
26
25
24
23
22
12
11
10
CIP
ADM
ADM
PWR
21
20
AST
19
18
17
16
PWR
RTR
MEN
SMD
SMC
SMB
SMA
SMD
SMC
SMB
SMA
18
17
16
PPD
PPD
PPD
PRS
PEN
PFL
PWR
CIP
ADM
AST
SMA
SMB
SMC
SMD
RTR
MEN
PWM_PERIOD
PWM_RESET
PWM_ENABLE
PWM_FIFO_LEVEL
Power on reset
Current Direction Input Polarity:
A2D multiplex
A2D Converter Start
State Machine A enable
State Machine B enable
State Machine C enable
State Machine D enable
Reset trip button state
Enable button state
DPR1 0xA0000100;
31
30
29
28
27
26
25
Read
Write
15
14
13
12
11
10
Read
Write
UI7
UI6
UI5
UI4
UI3
UI2
UI1
24
23
22
21
20
19
ABY
PVE
PVT
PAF
PAE
PEM
PFU
CINA
CINB
CINC
CIND
ABY
UI1-7
PVE
8
UI0
CIND
CINC
CINB
CINA
PFU
PEM
PAE
PAF
PVT
PWM_VECTOR
PWM_VECTOR_TIME
PWM_FIFO_Almost Full
PWM_FIFO_Almost Empty
PWM_FIFO_EMPTY
PWM_FIFO_FULL
Current Direction Input, PhaseA
Current Direction Input, PhaseB
Current Direction Input, PhaseC
Current Direction Input, PhaseD
A2D converters Busy
User input 1 to 7
DPR2 0xA0000200;
31
30
Read
Write
29
28
27
26
25
24
23
22
21
20
19
T3
T3
15
14
18
17
16
T2
T2
13
12
11
10
Read
Write
T1
T1
T1
T2
T3
DPR3 0xA0000300;
31
Read
Write
Read
Write
30
29
28
27
26
25
24
23
22
21
20
19
18
17
16
14
13
12
11
10
AD0
AD1
15
AD0
AD1
DPR4 0xA0000400;
31
Read
Write
Read
Write
30
29
28
27
26
25
24
23
22
21
20
19
18
17
16
14
13
12
11
10
AD3
AD2
15
AD2
AD3
DPR5 0xA0000500;
31
Read
Write
Read
Write
30
29
28
27
26
25
24
23
22
21
20
19
18
17
16
14
13
12
11
10
AD5
AD4
15
AD4
AD5
DPR6 0xA0000600;
31
Read
Write
Read
Write
30
29
28
27
26
25
24
23
22
21
20
19
18
17
16
14
13
12
11
10
AD7
AD6
15
AD6
AD7
DPR7 0xA0000700;
31
Read
Write
Read
Write
30
29
28
27
26
25
24
23
22
21
20
19
18
17
16
14
13
12
11
10
AD9
AD8
15
AD8
AD9
DPR8 0xA0000800;
31
30
29
28
27
26
25
24
23
22
21
20
19
18
17
16
Read
Write
HT23
HT22
HT21
HT20
HT19
HT18
HT17
HT16
HT15
HT14
HT13
HT12
HT11
HT10
HT9
HT8
15
14
13
12
11
10
Read
Write
HT7
HT6
HT5
HT4
HT3
HT2
HT1
HT0
ST7
ST6
ST5
ST4
ST3
ST2
ST1
ST0
ST7
ST6
ST5
ST4
ST3
ST2
ST1
ST0
22
21
20
19
18
17
16
ST0 ST7
HT0 HT23
Software Trip
Hardware Trip
Active High
Active High
HT0
HT1
HT2
HT3
HT4
HT5
HT6
HT7
HT8
HT9
HT10
HT11
HT12
HT13
HT14
HT15
DPR9 0xA0000900;
31
30
29
28
Read
Write
UO3
UO2
UO1
UO0
27
WEN
UO3
UO2
UO1
UO0
WEN
WSR
15
14
13
12
Read
Write
W_PERIOD
11
26
10
25
24
23
W_PERIOD
W_PERIOD
WSR
WEN
UO0-3
DPR10 0xA0000A00;
31
30
29
28
27
26
25
24
23
22
21
20
19
18
Read
Write
17
16
Address
15
14
Read
Write
13
12
11
10
Command
Register
I2C
Data
I2C DATA for controlling the Hardware trip limit setting via I2C digital potentiometers(MAX5478)
DPR12 0xA0000C00;
31
30
29
28
27
26
25
24
23
22
21
20
19
18
17
16
Read
Write
TE23
TE22
TE21
TE20
TE19
TE18
TE17
TE16
TE15
TE14
TE13
TE12
TE11
TE10
TE9
TE8
TE23
TE22
TE21
TE20
TE19
TE18
TE17
TE16
TE15
TE14
TE13
TE12
TE11
TE10
TE9
TE8
15
14
13
12
11
10
Read
Write
TE7
TE6
TE5
TE4
TE3
TE2
TE1
TE0
TE7
TE6
TE5
TE4
TE3
TE2
TE1
TE0
References
[1] D. Katsis, P.W. Wheeler, J.C. Clare, and P. Zanchetta. A three-phase
utility power supply based on the matrix converter. In Proc. IEEE
IAS Annual Meeting 04, Seattle, 2004.
[2] R. Zhang, D. Boroyevich, V.H. Prasad, H.-C. Mao, F.C. Lee, and
S. Dubovsky. A three-phase inverter with a neutral leg with space
vector modulation. In Proc. IEEE APEC97, volume 2, pages 857
863, February 1997.
[3] Domenico Casadei, Giovanni Serra, and Angelo Tani. Reduction
of the input current harmonic content in matrix converters under
input/output unbalance. IEEE Trans. Ind. Electron., 45(3):401411,
June 1998.
[4] D. Casadei, G. Serra, A. Tani, and L. Zarri. Matrix converter
modulation strategies: a new general approach based on spacevector representation of the switch state. IEEE Trans. Ind. Electron.,
49(2):370381, April 2002.
[5] Alberto Alesina and Marco G. B. Venturini. Solid-state conversion:
A fourier analysis approach to generalized transformer synthesis.
IEEE Trans. Circuits Syst., 28(4):319330, April 1981.
[6] Alberto Alesina and Marco G. B. Venturini. Analysis and design
of optimum-amplitude nine-switch direct ac-ac converters. IEEE
Trans. Power Electron., 4(1):101112, January 1989.
[7] L. Huber and D. Borojevic. Space vector modulator for forced
commutated cycloconverters. In Proc. IEEE IAS Annual Meeting 89,
San Diego, CA, 1989.
[8] L. Huber and D. Borojevic. Space vector modulation with unity
input power factor for forced commutated cycloconverters. In Proc.
IEEE IAS Annual Meeting 91, 1991.
253
R EFERENCES
[9] L. Huber, D. Borojevic, and N. Burany.
Analysis, design
and implementation of the space-vector modulator for forcedcommutated cycloconvertors.
Electric Power Applications, IEE
Proceedings B, 139(2):103113, March 1992.
[10] L. Huber and D. Borojevic.
Space vector modulated threephase to three-phase matrix converter with input power factor
correction. Industry Applications, IEEE Transactions on, 31(6):1234
1246, November 1995.
[11] M.J. Ryan, R.D. Lorenz, and R. De Doncker. Modeling of multileg
sine-wave inverters: a geometric approach. IEEE Trans. Ind.
Electron., 46(6):11831191, December 1999.
[12] M.J. Ryan, R.W. De Doncker, and R.D. Lorenz. Decoupled control of
a four-leg inverter via a new 4 times;4 transformation matrix. IEEE
Trans. Power Electron., 16(5):694701, September 2001.
[13] V.H. Prasad, D. Borojevic, and R. Zhang. Analysis and comparison
of space vector modulation schemes for a four-leg voltage source
inverter. In Applied Power Electronics Conference and Exposition, 1997.
APEC 97 Conference Proceedings 1997., Twelfth Annual, volume 2,
pages 864871, February 1997.
[14] R. Zhang, V.H. Prasad, D. Boroyevich, and F.C. Lee. Threedimensional space vector modulation for four-leg voltage-source
converters. IEEE Trans. Power Electron., 17(3):314326, May 2002.
[15] R. A. Van Eck. Frequency-changer systems using the cycloconverter
S
principle. IEEE Trans. Appl. Ind., 82(66):163 A 168, May 1963.
[16] Louis J. Lawson. The practical cycloconverter. IEEE Trans. Ind. Gen.
Appl., 4(2):141144, March 1968.
[17] Muhammad H. Rashid. Power Electronics Handbook: Devices, Circuits
and Applications. Academic Press, 2nd edition, 2006.
[18] Bin Wu, J. Pontt, J. Rodriguez, S. Bernet, and S. Kouro. Currentsource converter and cycloconverter topologies for industrial
medium-voltage drives. Industrial Electronics, IEEE Transactions on,
55(7):27862797, July 2008.
[19] R. Hagmann. Ac-cycloconverter drives for cold and hot rolling mill
applications. In Industry Applications Society Annual Meeting, 1991.,
Conference Record of the 1991 IEEE, pages 11341140, September
1991.
254
R EFERENCES
[20] Alberto Alesina and Marco G. B. Venturini. Intrinsic amplitude
limits and optimum design of 9-switches direct pwm ac-ac
converters. In Proc. IEEE PESC88, Kyoto, Japan, 1988.
[21] J. Oyama, T. Higuchi, E. Yamada, T. Koga, and T. Lipo. New control
strategy for matrix converter. In Proc. IEEE PESC89, Milwaukee,
WI, 1989.
[22] L. Wei and T.A. Lipo. A novel matrix converter topology with
simple commutation. In Industry Applications Conference, 2001.
Thirty-Sixth IAS Annual Meeting. Conference Record of the 2001 IEEE,
volume 3, pages 17491754, September 2001.
[23] C. Klumpner, F. Blaabjerg, I. Boldea, and P. Nielsen. A new
modulation method for matrix converters. In Industry Applications
Conference, 2001. Thirty-Sixth IAS Annual Meeting. Conference Record
of the 2001 IEEE, volume 4, pages 21432150, September 2001.
[24] Zhao Yu and Yan Yangguang. A three-phase four-line soft switched
inverter for ups applications.
In Industrial Electronics, 1992.,
Proceedings of the IEEE International Symposium on, volume 2, pages
696700, May 1992.
[25] G. Oriti, A.L. Julian, and T.A. Lipo. A new space vector modulation
strategy for common mode voltage reduction [in pwm invertors]. In
Power Electronics Specialists Conference, 1997. PESC 97 Record., 28th
Annual IEEE, volume 2, pages 1541 1546, June 1997.
[26] S.M. Ali and M.P. Kazmierkowski. Pwm voltage and current control
of four-leg vsi. In Industrial Electronics, 1998. Proceedings. ISIE 98.
IEEE International Symposium on, volume 1, pages 196201, July
1998.
[27] S. Ei-Barbari and W. Hofmann. Digital control of a four leg
inverter for standalone photovoltaic systems with unbalanced load.
In Industrial Electronics Society, 2000. IECON 2000. 26th Annual
Confjerence of the IEEE, volume 1, pages 729734, 2000.
[28] M.E. Fraser, C.D. Manning, and B.M. Wells. Transformerless
four-wire pwm rectier and its application in ac-dc-ac converters.
Electric Power Applications, IEE Proceedings -, 142(6):410416,
November 1995.
[29] D. Katsis, P.W. Wheeler, J.C. Clare, L. Empringham, and M. Bland. A
utility power supply based on a four-output leg matrix converter. In
255
R EFERENCES
Industry Applications Conference, 2005. Fourtieth IAS Annual Meeting.
Conference Record of the 2005, volume 4, pages 2355 2359 Vol. 4,
October 2005.
[30] P. W. Wheeler, P. Zanchetta, J. C. Clare, L. Empringham, M. Bland,
and D. Katsis. A utility power supply based on a four-output leg
matrix converter. IEEE Trans. Ind. Appl., 44(1):174186, February
2008.
[31] Jang-Hwan Kim and Seung-Ki Sul. A carrier-based pwm method
for three-phase four-leg voltage source converters. IEEE Trans.
Power Electron., 19(1):6675, January 2004.
[32] Pw. Wheeler, J.C. Clare, and N. Mason. Space vector modulation for
a 4-leg matrix converter. In Power Electronics Specialists Conference,
2005. PESC 05. IEEE 36th, pages 3138, June 2005.
[33] L. Empringham, P.W. Wheeler, and J.C. Clare.
Intelligent
commutation of matrix converter bi-directional switch cells using
novel gate drive techniques.
In Power Electronics Specialists
Conference, 1998. PESC 98 Record. 29th Annual IEEE, volume 1, pages
707713, 1998.
[34] Fan Yue, P.W. Wheeler, and J.C. Clare. A novel four-leg matrix
converter. In IEEE Industrial Electronics, IECON 2006 - 32nd Annual
Conference on, pages 2694 2699, November 2006.
[35] Fan Yue, P.W. Wheeler, N. Mason, L. Empringham, and J.C. Clare. A
new control method of single-stage 4-leg matrix converter. In Power
Electronics and Applications, 2007 European Conference on, pages 1 10,
September 2007.
[36] R. CaA andrdenas, R. Pe anda, P. Wheeler, and J. Clare.
s
Experimental validation of a space-vector-modulation algorithm
for four-leg matrix converters.
Industrial Electronics, IEEE
Transactions on, 58(4):12821293, April 2011.
[37] Sangshin Kwak, Taehyung Kim, and O. Vodyakho. Four-leg based
matrix converter with fault resilient structures and controls for
electric vehicle and propulsion systems. In Vehicle Power and
Propulsion Conference, 2007. VPPC 2007. IEEE, pages 519 523,
September 2007.
[38] Yao Sun, Mei Su, Lixun Xia, and Weihua Gui. Randomized carrier
modulation for four-leg matrix converter based on optimal markov
256
R EFERENCES
chain. In Industrial Technology, 2008. ICIT 2008. IEEE International
Conference on, pages 1 6, April 2008.
[39] Wesam M. Rohouma, Saul Lopez Arevalo, Pericle Zanchetta,
and PatrickW Wheeler. Repetitive control for a four leg matrix
converter. In Power Electronics, Machines and Drives (PEMD 2010),
5th IET International Conference on, pages 1 6, April 2010. doi:
10.1049/cp.2010.0208.
[40] S. Khwan-on, L. de Lillo, P. Wheeler, and L. Empringham. Fault
tolerant four-leg matrix converter drive topologies for aerospace
applications. In Industrial Electronics (ISIE), 2010 IEEE International
Symposium on, pages 21662171, July 2010.
[41] Y. Sun, M. Su, G. Weihua, H. Wang, and X. Li. Indirect four-leg
matrix converter based on robust adaptive back-stepping control.
Industrial Electronics, IEEE Transactions on, PP(99):1, 2010.
257